KR20200131750A - Sample data analysis method based on kernel modules in genomic module network and analysis apparatus - Google Patents

Abstract

The present invention relates to a sample analysis method based on a kernel module in a genome module network. The sample analysis method based on a kernel module in a genome module network comprises the steps of: allowing an analysis device to construct a genome module network for a sample based on entropy using gene expression data of the sample; and allowing the analysis device to perform analysis for the sample based on a kernel module in a reference genome module network and the kernel module in the genome module network of the sample. The kernel module is a module having the entropy which is equal to or lower than a reference value compared to other modules in the genome module network, and the entropy indicates a correlation between a plurality of genes based on possibility of transcriptional expression for the plurality of genes.

Description

Sample analysis method and analysis device based on the kernel module of the genome module network {SAMPLE DATA ANALYSIS METHOD BASED ON KERNEL MODULES IN GENOMIC MODULE NETWORK AND ANALYSIS APPARATUS}

The technique to be described below relates to a technique of analyzing a sample centering on the kernel module of the genome module network.

Traditionally, it is assumed that diseases such as malignancies are caused by the genome, so research to overcome malignancies is focused on the genome. The development of molecular biology has made it possible to reduce the side effects of traditional anticancer chemotherapy and to selectively destroy cancer cells. However, complete treatment for malignant tumors is still impossible. This is due to a lack of understanding of the function and mechanism of action of the genome. Traditional methods of research on the genome are based on biochemical techniques and have limitations in extending their understanding beyond chemical functions.

U.S. Patent No. 9,092,391

Conventional molecular biology tumor diagnosis predicts the likelihood of tumor development by searching for mutations in genes. However, it has not yet been established that mutation is the cause of tumor development, and there are no gene mutations applicable to all tumors. On the other hand, pathological tumor diagnosis is based on the morphological characteristics of tumor cells. Therefore, in terms of predicting tumor incidence, its effectiveness is diminished.

The technique described below is intended to provide a technique for performing analysis on a sample centering on a kernel module of a genome module network constructed using a gene expression data set.

In the sample analysis method based on the kernel module of the genome module network, the analysis device constructs a genome module network for the sample based on entropy using gene expression data of the sample, and the analysis device is a reference kernel of the reference genome module network. And performing analysis on the sample based on the module and the sample kernel module of the genome module network of the sample.

In another aspect, the sample analysis method based on the kernel module of the genome module network includes the steps of constructing a genome module network based on entropy using gene expression data obtained by combining the reference gene expression data and the sample gene expression data by an analysis device. And performing, by the analysis device, analyzing the sample based on the kernel module of the genome module network.

The reference genome module network is constructed in advance using at least one of a gene expression data set of a normal tissue and a gene expression data set of a tumor tissue, and the kernel module has an entropy of a reference value compared to other modules in the genome module network. It is a module that is abnormally low, and the entropy indicates a correlation between the plurality of genes based on the possibility of transcriptional expression for the plurality of genes.

In another aspect, the analysis device includes an input device that receives reference data and gene expression data of a sample, a storage device that stores a program for analyzing data based on a kernel module of a genome module network constructed as a gene expression data set, and the And a computing device for constructing a genome module network using the gene expression data of the sample using a program, and analyzing the sample based on gene information constituting the kernel module of the constructed genome module network.

The data of the reference is at least one of a gene expression data set of a normal tissue and a gene expression data set of a tumor tissue, or a reference genome module network data constructed using the at least one set, and the kernel module is the genome module In a network, an entropy is a module that is lower than a reference value compared to other modules, and the entropy indicates a correlation between the plurality of genes based on the possibility of transcriptional expression for the plurality of genes.

The technology to be described below identifies the difference between a normal tissue and a tumor tissue, centering on the kernel module of the genome module network, and provides a biological analysis or diagnosis for a sample based on this.

1 is an example of a genomic module network.
2 is an example showing the basic state of the genome.
3 is an example of a density matrix in genomic space.
4 is an example of a genome module located in the basic state matrix of the genome.
5 is an example of measuring gene expression in a genome space in a sample space.
6 is an example of a density matrix for an arbitrary module in a sample space.
7 is an example showing module perturbation due to the exclusion of gene i in the gene network of any module.
8 is an example of configuring a module-to-module network with TCGA data sets for eight organizations.
9 is an example of a network between modules of BRNO to which various cutoff values are applied.
10 shows an example of mapping the BRNO's inter-module network to modules of other organizations.
11 is an example showing kernel modules for eight organizations.
12 is an example showing the result of mapping a kernel module of BRNO to a kernel module of another organization.
13 is an example showing the result of mapping the module of the CCDR domain of BRNO to the module of another tissue.
14 is an example of a flow chart for the process of constructing a genome module network.
15 is an example of a process of calculating an analysis index for sample data using a genome module network.
Fig. 16 shows an example in which the sample probability (SP) of a plurality of tumor tissue samples is calculated using a genome module of a normal tissue.
17 shows an example of comparing tumor tissue sample groups classified based on SP.
18 is an example of classifying a sample of a tumor tissue based on a module sample probability (MSP).
19 is an example of displaying the average MSP for each module of a tumor tissue sample group in a genome module network of a normal tissue.
FIG. 20 is an example of defining a density matrix of a specific gene group in a gene space and showing the probability of a sample to be analyzed for the gene set.
21 is an example showing the log odds ratio (LOR) of a corresponding gene calculated in each sample of a tumor tissue using a density matrix of a specific gene group in a normal tissue.
22 is an example of the LOR of a gene calculated in each sample of a tumor tissue using a genome module of a normal tissue.
23 is an example of a flow chart for a method of analyzing sample data based on a filtering-based genome module network.
24 is an example of a process of calculating an analysis index for sample data using a filtering-based genome module network.
25 is an example of a filtering process for gene expression data.
26 is an example of a genome module network generated using data obtained by filtering BRCA.
FIG. 27 is an example of classifying samples based on MSPs for specific modules of the genome module network of FIG. 26.
28 is another example of classifying samples based on MSPs for a specific module of the genomic module network of FIG. 26.
29 is a survival curve for a specific sample group.
30 is a result of clustering based on the relative entropy of 8 tissues for CGX.
31 shows the results of clustering based on the relative entropy of 8 tissues for CG.
FIG. 32 is a result of calculating MSPCG and MSPCGX in a BRCA sample and showing the distribution of MSPCG and MSPCGX according to the expression of the receptor.
33 shows the results of clustering BRNO samples based on MSPCG and MSPCGX.
FIG. 34 is an example showing the relative entropy of each genomic module with respect to the kernel module in the BRNO inter-module network.
FIG. 35 is an example of displaying the relative entropy of each genomic module with respect to a kernel module based on a part of a sample in the BRNO inter-module network.
FIG. 36 is an example of displaying the relative entropy of each genomic module to CGX of a kernel module based on a part of a sample in the BRNO inter-module network.
37 is an example showing the relative entropy of each genomic module with respect to the CG of the kernel module based on a part of the sample in the BRNO inter-module network.
38 shows the results of clustering based on the MSP of the BRNO sample and the BRCA sample calculated for CG and CGX of BRNO.
39 is an example of a level plot clustered based on MSP of samples of BRNO and BRCA calculated for all BRNO genome modules.
FIG. 40 is an example showing a module having the lowest MSP among the MSPs of BRNO sample 12 for all BRNO genome modules in a network between BRNO modules.
41 is an example of a level plot clustered based on the relative entropy of the BN2211 module for BRNO modules mapped from BRNO to BN2211.
42 is the distribution of MSP of BRNO samples for modules 21, 30, 39 and 81 of BN2211.
43 is an example of a level plot clustered based on the relative entropy of BNRF modules for BN2211 modules mapped from BN2211 to BNRF.
44 is a result of calculating LOR for genes of modules 21, 30, 39, and 81 of BN2211.
FIG. 45 is an example of a level plot in which BRCA modules mapped from BRCA to BN2211 are clustered based on the relative entropy of modules of BN2211.
46 is an example of a level plot clustered based on the relative entropy of BRCA modules for BN2211 modules mapped from BN2211 to BRCA.
47 is a result of clustering based on the MSP of BRCA samples for some modules of BN2211.
48 is a survival curve for a group of BRCA samples classified by the criteria of MSPCG and MSPCGX.
49 is an example of a level plot clustered based on the relative entropy of BAHL modules for modules of BRNO mapped from BRNO to BAHL.
50 is an example showing the MSP distribution of BRCA samples for the BAHL module.
51 is an example showing the MSP of BRNO samples for BAHL modules 26, 32, 43, 53 and 60.
52 shows the results of clustering based on the MSP of BRCA samples for BAHL modules 26, 32, 43, 53 and 60.
53 is a result of clustering BRCA samples according to MSPCG and MSPCGX.
Figure 54 shows the results of a linear regression analysis between the median incidence of mutations in the BRCA sample group and the relative entropy of CG and CGX of BRCA to CG and CGX of BRNO.
55 is an example of a linear regression analysis result on the mutation frequency of a BRCA sample group in each module of BRNO in a network between modules of BRNO.
Figure 56 shows the relative entropy for the BRNO module of the 4 BRCA sample groups sorted according to mutation frequency.
Fig. 57 is a result of prior regression analysis of mutation frequency and relative entropy of BRCA for BRNO modules 43 and 51.
58 is an example of a process of analyzing a kernel module of a genome module network.
59 is an example of a process of analyzing a sample based on a kernel module of a genome module network.
60 is an example of a system for analyzing a sample based on a kernel module of a genome module network.
61 is an example of an analysis device that analyzes a sample based on a kernel module of a genome module network.

The technology to be described below may be modified in various ways and may have various embodiments, and specific embodiments will be illustrated in the drawings and described in detail. However, this is not intended to limit the technology to be described below with respect to a specific embodiment, and it should be understood to include all changes, equivalents, and substitutes included in the spirit and scope of the technology described below.

Terms used in the following description are first described.

A sample can basically mean a living entity. Entity basically means including humans, animals, plants, microorganisms, etc. However, in the following description, it is assumed that the sample is intended for humans. The sample may be represented by a sample obtained from an individual to be analyzed. Thus, the sample includes meanings such as an individual, an individual's tissue, an individual's cell collection, and the like.

The sample data refers to the gene expression data of the sample.

Gene expression means that a gene is transcribed into an RNA product.

Gene expression data is a data set representing the level of expression of genes. Meanwhile, gene expression data may be calculated using techniques such as microarray and NGS (Next Generation Sequencing).

A genomic module or module corresponds to a group of genes in the genome of higher multicellular eukaryotes. One genomic module consists of a plurality of genes.

Modularization is a process of constructing a plurality of genomic modules using a gene expression data set.

The intermodular network refers to a network in which a plurality of genomic modules are connected by edges. In the graph data structure, a genomic module corresponds to a node, and an edge refers to a link connecting nodes.

Edge refers to the connections between genomic modules in an inter-module network. The edge can be said to be a channel for exchanging or transferring information between genomic modules.

Genetic network refers to a network in which genes constituting a genomic module are connected by edges. A gene network is a network of genes inside a genomic module.

The genomic module network is generally meant to include inter-module networks and gene networks.

A domain or a genomic module domain means a specific region composed of a plurality of genomic modules in an inter-module network. As will be described later, the domain may be composed of modules having a biological common function.

Mapping is performed by overwriting data on a plurality of genes belonging to a first genome module network composed of a first gene expression data set or a specific module of a first genome module network to a second gene expression data set, and 2 It refers to an operation that performs a certain analysis based on a gene expression data set. In other words, some data of the gene expression data set (second gene expression data set) to be currently analyzed is extracted and analyzed from another gene expression data set (first gene expression data set). Here, a certain analysis may mean entropy operation, genomic module network reconstruction, and the like.

The genome space refers to the Hilbert space, which has a vector of the basis state of the genome as a coordinate axis.

The sample space refers to an m-dimensional space having m samples as coordinate axes in a given data set.

The sample probability (hereinafter referred to as SP) refers to the probability of each sample for a plurality of genes in the whole genome module network. For example, the sample probability may correspond to a quantified value representing the degree of variation of a specific sample based on a normal tissue for all genes of the sample.

The module sample probability (Modular Sample Probability, hereinafter referred to as MSP) refers to a sample probability for a plurality of genes belonging to a specific genomic module.

The domain sample probability (hereinafter referred to as DSP) refers to a sample probability for a plurality of genes belonging to a specific domain.

Log Odds Ratio (hereinafter referred to as LOR) is based on a specific gene and is based on the ratio of the first probability when the gene is in the genome module and the second probability when the gene is not in the genome module. Is the logarithmic value. LOR can refer to the effect of a specific gene on a genomic module (genome system).

LOR _MSP is the negative logarithm of the ratio of the MSP of a specific genomic module if the gene is present in the sample and the MSP of that genomic module if the gene is not present in the sample, based on a specific gene in a specific sample. LOR _MSP can indicate the effect of certain genes on certain genomic modules.

LOR _DSP is the negative logarithm of the ratio of the DSP of a specific domain when the gene is present in the sample and the DSP of the domain when the gene is not present in the sample, based on a specific gene in a specific sample. LOR _DSP can indicate the effect of a specific gene on a specific domain.

LOR _SP is a negative logarithmic value of the ratio of the SP of the sample when the gene is present in the sample and the SP of the sample when the gene is not present in the sample, based on a specific gene in a specific sample. LOR _SP can indicate the effect of a specific gene on a specific sample.

The principal eigenvector refers to an eigenvector having the largest eigenvalue as a result of Singular Value Decomposition (SVD).

The kernel module or kernel refers to a module(s) having a lower entropy compared to other modules, not only in the original organization but also in other types of organizations in which mapping has been performed.

Entropy refers to the degree of functional unity and characteristic activity of multiple genes.

Construction of the genomic module network and analysis based on the genomic module network can be performed on a computer device. The computer device refers to a device capable of processing or processing input data uniformly. For example, the computer device may be any one of devices such as a PC, a smart phone, a server, and a chipset in which a program is embedded. Construction of the genomic module network and the analysis based on the constructed genomic module network can be performed in one single device. Alternatively, the construction of the genome module network and the analysis based on the constructed genome module network may be performed in a separate device. Hereinafter, a computer device for constructing a genome module network and/or performing an analysis based on the constructed genome module network is referred to as an analysis device.

The technique described below can clarify the relationship between the genomic module network and the phenotype. Researchers explore the flow of information about the genome's transcriptional activity of a gene, and confirm it in terms of phenotype. The genome module network can be utilized in various aspects. Researchers can build a genome module network using gene expression data for a specific sample, and analyze a sample based on the constructed genome module network. For example, analysis of the sample may include items such as disease onset, disease onset probability, survival rate, survival period, customized treatment method, and the like.

Most of the conventional studies involving certain diseases (eg, malignant tumors) are based on biochemical techniques. The technology described below has a completely different perspective from the biochemical technique, and corresponds to a technique that sees living organisms as a system and analyzes them.

Living organisms evolved from nucleated cells to nucleated cells, and from single cells to multicellular, vertically and horizontally complex structures. Vertically, a multilayered structure was formed, and horizontally, it developed into an evolved biological system by forming complex connections between multiple components. In general, a system is a collection of components connected to each other in an organized way, and one or a set of components affects the properties of the system. Ackhoff (1972) and Checkland (1981) stated that a system expresses the characteristics of the system itself rather than the characteristics of individual components or parts. According to the same principle, it can be said that the characteristics of the system itself are expressed, not the characteristics of proteins and genes, which are constituents, and the characteristics of the biological system itself become phenotype. Proteins and genes can affect the phenotype of living organisms, but they are difficult to regard as the building blocks of living systems themselves.

The biological system responds by expressing an appropriate phenotype to changes in the internal and external environment. The only place where this response scenario can be encoded is the DNA chain. Therefore, it can be said that the entire biological system is characterized by information on genes. The technique described below constructs a genome module network using a gene expression data set of a sample, and analyzes a sample through a system called a genome module network.

Construction of genome module network

First, the process of constructing the genome module network will be described.

1 is an example of a genome module network 100. Before describing the process of constructing the genome module network 100, the structure of the genome module network 100 will be described first. The genome module network 100 includes an inter-module network and a gene network. The module-to-module network includes a plurality of genomic modules and edges connecting the modules.

Each module of the genomic module network 100 contains a certain gene. In FIG. 1, the module is shown as a solid circle. The number indicated on the module is an example of an identifier that identifies the module. Genes belonging to one module correspond to genes associated with specific phenotypic expression. Meanwhile, a plurality of modules may be divided into groups that perform a certain function in relation to the phenotype. In Fig. 1, a plurality of modules are divided by a dotted circle. The dotted circle is an example of dividing a plurality of modules into related functions. As such, a domain is an area in which a plurality of functionally related modules are divided. In Figure 1, A, B, C, D and E domains are indicated. Meanwhile, module 84 serves as a relay linking domain A and domain C. Each domain may be associated with functions such as cell cycle, DNA damage control, epithelial tissue, extracellular matrix, immunity, angiogenesis, and the like.

1 shows a state in which a pair of modules are connected by edges. It means that the two modules connected by edges have a certain relationship with each other. Here, the edge can be seen as a channel for transferring or exchanging information between modules. For modules connected by edge, one module can be considered to be related to the function of another module. When modules belonging to different domains are connected, it can be interpreted that one domain affects the function of the other domain. For example, a specific phenotype may be expressed through one module, but various modules may be directly or indirectly involved and expressed.

The module 27 is enlarged at the bottom of FIG. 1. As described above, each module is composed of a plurality of genes. Genes belonging to module 27 were identified by alphabet. Genes belonging to one module constitute a gene network. The gene network is a form in which genes within a module are connected by edges, similar to forming a network between modules.

Hereinafter, the process of building a genome module network includes a process of modularizing a genome, a process of building a network between genome modules, and a process of building a gene network within each module. Each process will be described below.

State of genome

The concept of defining the state of the gene will be explained. The state of the gene is described at the level of the quantum system. Quantum systems are expressed in the form of a density matrix.

및

으로 표현한다. 하나의 유전자의 전사 상태 벡터(transcriptional state vector)

는 아래의 수학식 1과 같이 두 개 기초 상태의 선형적 조합에 해당한다. One gene has two basis states: active to inactive. The active state means that the gene is active during the transcription process. At a certain point in time, a gene has an active or inactive state. Therefore, the basic states are mutually exclusive and are mathematically orthonormal in vector space. The active state may be expressed as "1" or "on", and the inactive state may be expressed as "0" or "off". For convenience of explanation, each of the active and inactive states is a basis state vector.

And

Expressed as Transcriptional state vector of one gene

Corresponds to a linear combination of two basic states as shown in Equation 1 below.

및

은 직교 정규(orthonormal) 특성을 갖는다. In Equation 1, a ₀ is a coefficient for an inactive state, and a ₁ is a coefficient for an active state. The amount of mRNA produced by the gene depends on a ₁ . Basic state vector

And

Has an orthonormal characteristic.

및

을 표시하였다.2 is an example showing the basic state of a gene. Figure 2 shows the basal state of two genes g ₁ and g ₂ . Both genes can have four basic states. Figure 2 is a transcriptional state vector for each of the genes g ₁ and g ₂

And

Is indicated.

The probability distribution of the states of genes in the genome indicates characteristic relationships between genes. If the genes have a uniform distribution state, the genes can be said to have random activities without being related to each other. However, as the association between genes increases, the unevenness of the state distribution of genes increases. Therefore, the non-uniformity of the probability distribution of the gene status can be said to be information indicating the degree of association of the gene in the genome.

로 나타낼 수 있다. 여기서 j _i ∈{0,1}이고, i = 1,2,...,n이다. 결국, n 개의 유전자로 구성된 게놈은 직교 정규 특성을 갖는 2ⁿ개의 기초 상태 벡터

를 갖는다. 유전자의 개수에 따라 새로운 벡터 공간이 늘어난다고 할 수 있다. 게놈의 기초 상태 벡터로 정의되는 공간은 힐베르트 공간이다. 게놈의 기초 상태 벡터로 정의되는 공간을 게놈 공간이라고 명명하였다. A genome with n genes has a total of 2 ⁿ basic states. Each of the n genes has a basic state vector with orthogonal normal characteristics. So the basic state vector of the genome is

It can be expressed as Where j _i ∈{0,1} and i = 1,2,...,n. After all, a genome consisting of n genes is 2 ⁿ basal state vectors with orthogonal canonical properties

Has. It can be said that the space for new vectors increases with the number of genes. The space defined as the basic state vector of the genome is the Hilbert space. The space defined as the basic state vector of the genome was called the genome space.

에서 α번째 기초 상태를

라고 한다.

는

로 표현된다. j _i,α ∈{0,1}이다. 모든

는 서로 직교 정규하는 특성을 갖는다. 따라서 유전자 i의 전사 상태 벡터는 아래의 수학식 2와 같이 표현할 수 있다.Full basal state

The α-th basic state at

It is called.

Is

Is expressed as j _i,α ∈{0,1}. all

Have characteristics that are orthogonal to each other. Therefore, the transcription state vector of gene i can be expressed as in Equation 2 below.

에 따른다. 게놈은 해당하는 유전자를 제어하여 mRNA 생성을 제어한다.The degree of mRNA production of gene i is counted

Follows. The genome controls mRNA production by controlling the corresponding gene.

를 유전자 i의 밀도 행렬(density matrix) ρ_i이라고 한다.

은

와 동일하기 때문에, 밀도 행렬은 게놈의 순수 상태(pure state)를 나타낸다. 퀀텀 시스템에서 순수 상태는 상태를 정확하게 알고 있는 상태를 의미한다. 게놈 시스템의 확률적 특성을 고려하면 밀도 행렬을 이용하여 유전자의 순수 상태의 조합(ensemble of pure states)으로 게놈의 혼합 상태(mixed state)를 설명하는 것이 유용하다. 따라서 ρ_i는 아래 수학식 3과 같이 나타낼 수 있다.Dyad normalized so that the trace is 1

Is called the density matrix ρ _i of gene i.

silver

The density matrix represents the pure state of the genome because it is equal to. In a quantum system, a pure state means a state that accurately knows the state. Considering the probabilistic characteristics of the genomic system, it is useful to describe the mixed state of the genome as an ensemble of pure states using a density matrix. Therefore, ρ _i can be expressed as Equation 3 below.

이다. w_i는 ρ_i의 확률이다. w_i가 1/n으로 동일한 값이라면, ρ는

로 표현될 수 있다. 게놈 공간은 힐베르트 공간이므로, 밀도 행렬 ρ에 대한 단위 벡터

의 확률은 아래의 수학식 4와 같이 글리슨 정리(Gleason's theorem)에 따라 정의될 수 있다. Therefore, the mixed state density matrix ρ of the genome corresponds to the combination of ρ _i . That is, ρ is

to be. w _i is the probability of ρ _i . If w _i is equal to 1/n, then ρ is

It can be expressed as The genomic space is the Hilbert space, so the unit vector for the density matrix ρ

The probability of can be defined according to Gleason's theorem as shown in Equation 4 below.

번째 기초 상태에 머무를 확률은

이다. 게놈이 특정 기초 상태일 가능성은 게놈의 밀도 행렬의 대각에 위치한다. n 개의 유전자로 구성되는 게놈에 대한 밀도 행렬은 2ⁿ×2ⁿ 크기의 정방 행렬이다. 이 밀도 행렬은 2ⁿ 개의 고유 벡터(eigenvector)와 고유값(eigenvalue)을 갖는다. 고유 벡터는 고유상태(eigenstates)를 나타내고, 고유값은 특정 상태에 대한 확률을 나타낸다.The genome is

The probability of staying in the first basic state is

to be. The likelihood that a genome is in a certain underlying state lies diagonally to the density matrix of the genome. The density matrix for the genome consisting of n genes is a square matrix of size 2 ⁿ × 2 ⁿ . This density matrix has 2 ⁿ eigenvectors and eigenvalues. Eigenvectors represent eigenstates, and eigenvalues represent probabilities for specific states.

은 게놈의 기초 상태 벡터를 나타낸다. 실선 화살표로 표시된 2ⁿ개의 축

은 고유벡터를 나타낸다. 굵은색 화살표의 길이는 고유 벡터

의 확률을 나타낸다. 검은색 점은 유전자를 나타낸다.The probability of a genomic system staying in each native state is non-uniform. In genomic systems, this non-uniformity corresponds to certain genetic information. 3 is an example of a density matrix in genomic space. The density matrix has an elliptical shape in the two-dimensional genomic space. 2 ⁿ axes indicated by dashed arrows

Represents the basic state vector of the genome. 2 ⁿ axes indicated by solid arrows

Represents an eigenvector. The length of the bold arrow is an eigenvector

Represents the probability of Black dots represent genes.

의 고유 벡터는 게놈 시스템의 창발적 특성을 나타내고, 고유 벡터의 고유값은 특성 발현의 확률을 결정한다. 불균일성은 엔트로피 S(ρ)로 표현할 수 있다. 게놈이 어떤 상호 작용에서도 유전자를 활성화하지 않거나, 매우 많은 상호 작용에서 동시에 다수의 유전자를 활성화하면, 높은 엔트로피 값을 갖는다. 게놈 공간에서 엔트로피가 증가하면 밀도 행렬의 타원은 특정한 방향성을 잃고 원 형태를 갖게 된다. 반대로 게놈이 소수의 특정한 타깃에 집중하여 활동하면 낮은 엔트로피 값을 갖는다. 게놈 공간에서 생성된 유전 정보는 실제 공간(real space)에서 단백질 네트워크로 전달된다. mRNA는 게놈 공간과 단백질 공간을 연결하는 채널에 해당한다. Mixed state density matrix of genome

The eigenvector of represents the emergent characteristic of the genomic system, and the eigenvalue of the eigenvector determines the probability of characteristic expression. The non-uniformity can be expressed as entropy S(ρ). If the genome does not activate a gene in any interaction, or activates a large number of genes simultaneously in very many interactions, it has a high entropy value. As entropy increases in genomic space, the ellipse of the density matrix loses its specific orientation and takes on a circular shape. Conversely, if the genome is active by focusing on a small number of specific targets, it has a low entropy value. Genetic information generated in genomic space is transferred from real space to protein networks. mRNA corresponds to a channel that connects the genomic and protein spaces.

Genome modularization

Higher eukaryotes run different protein networks simultaneously in a single cell. It is assumed that genes involved in a specific interaction belong to a group. Genes belonging to certain groups work in association to produce proteins that represent phenotypes associated with specific interactions. Therefore, genes belonging to the group can be defined as one module. Modules are genes involved in the production of proteins involved in a specific phenotype. By analyzing the genes of the entire genome, the genome can be divided into a plurality of modules. Genes belonging to the module can be directly involved in producing proteins for a specific phenotype. Furthermore, genes belonging to the module may be indirectly involved in the process of producing a specific protein.

The researcher divides the whole genome into independent modules as much as possible, and analyzes the association between the independent modules to determine the linkage (edge) between modules. By constructing a genome module network for a specific genome, we want to analyze the genome at the genome module network level.

Multiple modules can be cooperatively involved in the expression of a particular phenotype. It can be said that a plurality of modules perform constant communication through the edge between modules.

In principle, if the gene index and the underlying state are properly aligned, the genome module can be separated. The probability of a genome staying in each basal state is mostly close to zero and will fluctuate in the genomic module region.

In the case of single-celled organisms, a single gene cannot have multiple states at the same time, so it can only play a role. In other words, because mRNA maintains one level in a physically contiguous space, one gene would have to be contained in one genomic module.

On the other hand, in multicellular organisms, a gene is expressed in each physically separated space. Similar to multitasking through time division of a computer central processing unit, a gene in a nucleated organism can perform multitasking through spatial division, which provides a basis for the evolution of a nucleated organism into a multicellular organism.

4 is an example of a genome module located in the basic state matrix of the genome. The vertical axis represents the index of the basal state, and the horizontal axis represents the gene index. Module c and module b partially overlap.

Modules a, b and d or modules a, c and d can be activated in one cell (single genomic space). However, modules b and c have shared genes and must be activated in different cells (multi-genome space).

Modules a and b partially overlap the basic state, but the eigenvectors of the two modules have different directions. Thus, the two modules a and b are involved in different protein networks and phenotypes. Mutual information I(a:b) for two modules a and b is expressed as S(ρ _a ) + S(ρ _b )-S(ρ _ab ). Mutual information refers to interdependence between two modules. When the basic state shared by the two modules increases, S(ρ _ab ) decreases, and mutual information increases. The number of underlying states shared between two modules can affect the degree of connectivity between genomic modules. However, when each genomic module has a degree of complexity enough to express its own characteristics, this linkage functions as a parameter to the execution of the genomic module.

Real space and genomic space are essentially different. The genome space is a 2 ^n- dimensional space in which the basic state of the genome is defined as a unit vector. The real space is a three-dimensional space in the real world where chemical reactions such as the production of specific proteins through gene activity occur in living organisms. Therefore, it is impossible to directly access the genomic space to determine the activity of the genome. Therefore, there is a need for a method for analyzing by converting the genomic space into a sample space of gene expression data. The sample space is an m-dimensional space in which each sample is defined as a unit vector.

Measurement of gene expression from m samples transforms the information carried on the mRNA in the genomic space into the m-dimensional sample space. On the other hand, technologies such as cDNA microarrays can simultaneously measure the expression levels of tens of thousands of genes.

는 전사 상태(transcriptional state) 행렬

를 구성한다. m 개의 샘플로부터 측정된 n 개의 유전자 발현 데이터 세트에서 유전자 i의 순수 상태 벡터

는 행렬

를 구성한다. 두 행렬의 관계는 아래의 수학식 5와 같다.Transcription state vector of gene i in a genome with n genes

Is the transcriptional state matrix

Configure. Pure state vector of gene i in n gene expression data sets measured from m samples

Is a matrix

Configure. The relationship between the two matrices is shown in Equation 5 below.

은 게놈 공간을 샘플 공간으로 변환하는 변환 행렬이다.here

Is a transformation matrix that transforms genomic space into sample space.

의 관계를 갖게 된다. 밀도 행렬 ρ는 실대칭행렬(real symmetric matrix)이므로 고유값 분해(Eigen Decomposition)를 하면

의 결과를 얻을 수 있다. 여기서 Q는 각 열(column)이 밀도 행렬 ρ의 각 고유 벡터인 직교 정규 행렬이며, Λ는 각 대각 성분이 밀도 행렬 ρ의 각 고유값인 대각 행렬이다. 따라서 수학식 5로부터 G ^T G를 전개하면

이 성립한다. From Equation 3 above, the whole genome density matrix ρ and the transcription state matrix T are

Will have a relationship. Since the density matrix ρ is a real symmetric matrix, the eigen decomposition

You can get the results. Here, Q is an orthogonal normal matrix in which each column is each eigenvector of the density matrix ρ, and Λ is a diagonal matrix in which each diagonal component is each eigenvalue of the density matrix ρ. Therefore, if G ^T G is developed from Equation 5

This holds true.

을 SVD로 분해하면

의 결과를 얻을 수 있다. 여기서

는 좌 특이벡터(left singular vector) 행렬이고,

는 특이값(singular value) 행렬이고,

는 우 특이벡터(right singular vector) 행렬이다. 따라서

이 성립한다. 여기서

로서, Q'는 게놈 전체의 밀도 행렬 ρ의 각 고유 벡터가 변환 행렬

의 좌 특이벡터 행렬을 통해 회전된 새로운 고유 벡터의 행렬임을 의미한다. Also the transformation matrix

Decomposing into SVD

You can get the results. here

Is the left singular vector matrix,

Is the singular value matrix,

Is a right singular vector matrix. therefore

This holds true. here

As, Q 'each are the eigenvectors of the density ρ of the genome matrix transformation matrix

It means that it is a matrix of new eigenvectors rotated through the left singular vector matrix of.

이 성립한다. On the other hand, if von Neumann entropy is invariant for unitary transformation,

This holds true.

엔트로피는

이 성립한다. 여기서 Q'ΛQ' ^T 는 밀도 행렬 ρ가 회전만 하고 고유값 행렬 Λ는 변하지 않은 상태이다. 또한 Λ의 대각 성분, 즉 ρ의 고유값은

이며, 엔트로피가 낮다면 고유값은 급격하게 감소하게 된다. 즉,

으로 가정할 수 있으며, Λ의 처음 m 개의 고유값만으로 대각행렬

을 생성하면,

이 성립한다. 여기서

는

를 구성하는 2ⁿ 개의 고유 벡터 중 처음 m 개의 고유 벡터만으로 구성된 행렬이다. In other words,

Entropy is

This holds true. Here, Q'ΛQ' ^T is a state in which the density matrix ρ is rotated and the eigenvalue matrix Λ remains unchanged. Also, the diagonal component of Λ , that is, the eigenvalue of ρ, is

If the entropy is low, the eigenvalue decreases rapidly. In other words,

Can be assumed, and only the first m eigenvalues of Λ

If you create

This holds true. here

Is

It is a matrix consisting of only the first m eigenvectors of 2 ⁿ eigenvectors constituting.

의 특이값 행렬

에서 처음 m 개의 행에서 대각 성분은 1이며 나머지는 0으로 채워진다. 따라서

가 성립한다. 여기서 W는 행렬

에서 처음 m 개의 행을 추출하여 생성한 행렬이다.If all samples are of the same kind of tissue, then the transformation matrix

Singular value matrix of

In the first m rows in, the diagonal component is 1 and the remainder is padded with 0s. therefore

Is established. Where W is the matrix

It is a matrix created by extracting the first m rows from.

이므로 Q'와 Q를 구성하는 열 벡터

와

사이에는

가 성립한다. 여기서 Q'는 Q를 U에 대해 회전한 행렬이므로 각 고유 벡터 간의 정규직교성(orthonormality)은 유지된다. 따라서

는 i=j일 때 1이고 i≠j일 때 0이며,

가 성립한다.

를 구성하는 2ⁿ 개의 특이벡터 중 처음 m 개의 특이벡터로 생성한 행렬

은 변환 행렬

의 가장 중요도가 높은 특이벡터로 구성되므로

가 성립한다. 여기서

은 행렬 W의 열벡터

와 같으므로

가 성립한다. 따라서 행렬 W는 직교정규행렬로 간주할 수 있다. Meanwhile,

Because columns that make up the Q 'and Q vector

Wow

In between

Is established. Here, Q'is a matrix in which Q is rotated about U , so the orthonormality between each eigenvector is maintained. therefore

Is 1 when i=j and 0 when i≠j,

Is established.

A matrix created with the first m singular vectors of 2 ⁿ singular vectors

Is the transformation matrix

Is composed of singular vectors with the highest importance of

Is established. here

Is the column vector of matrix W

Is the same as

Is established. Therefore, the matrix W can be regarded as an orthogonal normal matrix.

는 1에 근접하게 된다. 즉

는 밀도 행렬 ρ가 2ⁿ 차원의 게놈 공간에서 m 차원의 샘플 공간으로 변환된 밀도행렬로 간주할 수 있다. 따라서

의 엔트로피의 근사값은 아래의 수학식 6과 같다.Also, the sum of the diagonal components of Λ ' (the first m eigenvalues of Λ )

Comes close to 1. In other words

Can be regarded as a density matrix in which the density matrix ρ is transformed from 2 ^n- dimensional genomic space to m-dimensional sample space. therefore

The approximate value of the entropy of is shown in Equation 6 below.

이며 이는

과 근사하게 된다. 이를 종합하면 아래의 수학식 7을 얻을 수 있다.here

Which is

And it becomes cool. By synthesizing this, Equation 7 below can be obtained.

Equation 7 means that the entropy calculated from the data obtained by measuring gene expression is almost identical to the entropy of the transcriptional state of the genome.

이다. 그러나 게놈 시스템의 밀도 행렬 ρ와 유전자 i의 전사 상태를 직접 확인할 수 없고 단지 유전자 발현 데이터 G만을 알고 있을 뿐이다. 따라서 G로부터 확률을 계산할 수 있는 방법을 찾는 것은 중요한 과정이다. 먼저,

를 고유값분해하면

이 성립된다. 여기서

는 유전자 전사 상태 벡터

가 게놈의 밀도 행렬의 고유 벡터를 기저 벡터로 하는 좌표계로 변환된 벡터로서

로 치환할 수 있다. 따라서

가 성립한다. From Equation 4, the probability of gene i for the whole genome in genomic space is

to be. However, the density matrix ρ of the genomic system and the transcriptional status of gene i cannot be determined directly, only gene expression data G is known. Therefore, finding a way to calculate the probability from G is an important process. first,

Eigenvalue decomposition of

Is established. here

A gene transcription state vector

As a vector transformed into a coordinate system using the eigenvector of the density matrix of the genome as the basis vector,

Can be substituted with therefore

Is established.

는 아래의 수학식 8과 같다. Meanwhile, the pure state of the gene in the sample space

Is as shown in Equation 8 below.

에 대한 각 유전자의 확률은

로서, 수학식 5로부터

가 성립한다. 여기서

를 고유값분해

하고 변환 행렬

을 SVD로 분해(

)하여 적용하면

가 성립한다. 여기서 변환 행렬

의 특이벡터는 직교정규하므로

이고,

임을 적용하면

이 성립한다. 또한

는

가 변환 행렬

의 특이벡터

를 기저 벡터로 하는 좌표계로의 변환과

에 의한 차원 축소를 거친 벡터로서,

로 치환할 수 있다. 따라서 샘플 공간에서

에 대한 각 유전자의 확률은 아래의 수학식 9와 같다.In sample space

The probability of each gene for is

As, from Equation 5

Is established. here

Eigenvalue decomposition

And transformation matrix

To SVD (

) And apply

Is established. Where the transformation matrix

Since the singular vector of is orthogonal

ego,

If you apply

This holds true. In addition

Is

Transformation matrix

Singular vector of

Into a coordinate system using the basis vector

As a vector that has undergone dimension reduction by

Can be substituted with So in the sample space

The probability of each gene for is as shown in Equation 9 below.

이며 이는

과 근사하게 된다. 이를 종합하면 아래의 수학식 10을 얻을 수 있다. here

Which is

And it becomes cool. By synthesizing this, Equation 10 below can be obtained.

Equation 10 means that the probability of the gene calculated from the data obtained by measuring gene expression is almost identical to the probability calculated from the density matrix of the genome and the transcriptional state of the gene.

로부터 직접적으로 게놈 또는 유전자의 상태를 판단하려면

이 필요하다. Measured expression level

To determine the state of a genome or gene directly from

I need this.

은 많은 요인에 영향을 받을 수 있으며, 동일 조건이라 하더라도 각 유전자가 받는 영향은 모두 다르다. 결과적으로

는 실험 조건 내지 환경에 영향을 받는다. 이는 유전자 발현 데이터가 원칙적으로 일관성이 없음을 설명한다. 이러한 데이터의 취약성을 통계적 또는 실험적인 방법으로 극복하는 데에는 큰 한계가 있다.On the other hand, the selection of samples in time or sample space, measurement methods, and data processing affect the process of measuring gene expression. Thus, the transformation matrix

Can be affected by many factors, and even under the same conditions, the effects of each gene are all different. As a result

Is affected by experimental conditions or environment. This explains that gene expression data are in principle inconsistent. There is a great limitation in overcoming the vulnerability of these data by statistical or experimental methods.

로 표현되는 게놈 모듈에 대한 유전자 i의 확률은 게놈 공간에서의 확률

과 샘플 공간의 확률

이 일치한다. 나아가, 게놈 공간의 엔트로피

는 샘플 공간의 엔트로피

와 동일하게 된다. 이는 전술한 확률과 엔트로피가 유전자 발현 레벨의 차이에 따른 측정 환경에 영향을 받지 않는 파라미터에 해당한다는 것을 증명한다. 진핵생물의 게놈을 측정하기 위한 완벽한 변환 행렬

을 얻는 것은 불가능하지만, 엔트로피 및 확률은 측정 과정을 고려하지 않고 얻을 수 있는 것이다.The above-described Equations 4, 8, and 10 mean the following fact. Density matrix

The probability of gene i for a genomic module expressed as is the probability in genomic space

And the probability of sample space

Matches this. Furthermore, the entropy of genomic space

Is the entropy of the sample space

Will be the same as This proves that the above-described probability and entropy correspond to parameters that are not affected by the measurement environment according to the difference in gene expression level. Complete transformation matrix for measuring eukaryotic genomes

It is impossible to obtain, but entropy and probability can be obtained without considering the measurement process.

로 구성되는 기반의 2ⁿ 차원 게놈 공간에서 게놈 모듈의 밀도 행렬 ρ 및 유전자 전사 상태 벡터

가 기초 벡터

로 구성되는 m 차원의 샘플 공간에서 각각

및 유전자 발현 벡터

로 변환되는 예이다. 여기서

이므로 모듈에 대한 유전자의 확률은 유전자 발현의 측정에 따른 게놈 공간에서 샘플 공간으로의 변환에 영향을 받지 않는다.5 is an example of measuring gene expression in a genome space in a sample space. Foundation vector

Density matrix ρ of genome modules in a base 2 ^n- dimensional genomic space consisting of and gene transcription state vectors

Fall basis vector

Each in the m-dimensional sample space consisting of

And gene expression vectors

This is an example that is converted to. here

Therefore, the probability of a gene for a module is not affected by the transformation from genomic space to sample space according to the measurement of gene expression.

If the vectors of all genes in the sample space have the same orientation, the entropy is zero. This means that the elliptical density matrix is a straight line that matches the first eigenvector. If the probability of a gene is the same for all eigenvectors and the density matrix becomes a perfect circle (or sphere), then the entropy has a maximum value.

Hereinafter, an example of constructing the above-described genome module network using actual gene expression data will be described. A tumor is a small, independent system that resides in a large host system. Therefore, we intend to build a genome module network using genetic information about tumors.

Gene expression data sets include 6 types of tumor tissues, including breast cancer (BRCA), colon cancer (COAD), rectal cancer (READ), lung adenocarcinoma (LUAD), lung squamous cell cancer (LUSC), and ovarian cancer (OV), and normal breast tissue ( BRNO), normal colon tissue (CONO), data of 2 types of normal tissue, random mixture of 6 types of tumor tissue data (X6CA), random mix of 2 types of normal tissue data (X2NO), and 6 types of tumor tissue and 2 types Random mix of normal tissue data (X6C2N) data was used. BRCA and the like refer to a data set that is open to the public for academic research by measuring the amount of gene expression from the corresponding tissue in TCGA (The Cancer Genome Atlas). In order to reduce the computation time, each of 36 samples was randomly selected for BRCA, COAD, LUSC, and OV having more data sets than 36 samples. The genomic module was isolated using the data set.

와 같다. 따라서 전체 엔트로피는 각 모듈의 엔트로피의 합과 같게 된다. 즉

이다.The process of constructing the genome module based on the gene expression data set is described. Among the above, the contents necessary for modularization will be briefly described. In n modules that are completely independent (with no connectivity), each density matrix is ρ ₁ ,... It is called ρ _n . Each space is a Hilbert space, so the overall density matrix is

Same as Therefore, the total entropy is equal to the sum of the entropy of each module. In other words

to be.

이다. However, if n modules are not independent from each other (if there is connectivity between modules), the total entropy is smaller than the sum of the entropy of each module. In other words

to be.

One module affects other modules and exchanges certain information between modules. One module contains genes, and different modules may share the same gene. Therefore, it is difficult for each module to function completely independently in the genome module network. In this case, the criterion for module separation may be to find a module that minimizes the difference between the sum of the entropy of each module and the total entropy.

However, the analysis device not only does not know the actual number of modules that are activated in the genome, but also does not have information on the range of genes participating and genes participating in multiple modules simultaneously. Therefore, it is practically impossible to find a combination of modules using the above method. As a solution to this, the analysis device can grasp the local optimal points of the true module that actually exists and configure the module around them to complete the estimation module. In this process, the analysis device creates estimation modules that are close to the intrinsic module by preventing the transition to another local optimal point. The outer limits of the estimation modules, which are substantially overlapped with each other, coincide with the domain, the linking group of intrinsic modules that express phenotypes in a large category. This is because domains that control a large range of phenotypes are bound to have a small number of information exchange channels with other domains (max-flow min-cut).

Table 1 below is an example of a pseudo code for an algorithm for finding a local optimal point for a genomic module. Genes are divided into random sets, and genes are removed one by one for each divided set (module), and the local optimal point is found by lowering the entropy to the target value. Since it is necessary to find a local optimal point that exists inside the actual module, the entropy target value is set sufficiently low. In Table 1, "th" corresponds to the target value. In Table 1, the backslash'\' operation means an operation that removes the right element from the left set.

The genomic module is finally determined by using the local optimal point found through the process of Table 1. Table 2 below is an example of an algorithm for the process of completing the genome module.

Table 2 shows the expansion of the module by adding an external gene j one by one under conditions that do not increase entropy. In order to prevent the center of the module from moving during the module expansion process, the fluctuation in the direction of the feedback vector v ₁ is limited. In Table 2, "th" means a threshold value for the angle of variation of the reference vector.

At this time, since the optimal parameters such as the target value of entropy in Table 1 and the angle of fluctuation of the backing vector in Table 2 vary depending on the characteristics of the gene expression data, a consistent genomic module network construction and domain are identified from the results obtained with various parameters. The result is that it may be necessary to determine the optimal parameters.

In general, low entropy refers to a system focusing on a specific target represented by the first eigenvector of the density matrix. It can be said that genomic modules with low entropy in the genomic system of eukaryotic cells generate information for expressing a specific phenotype.

Some of the genes that make up the genomic module are duplicated in different modules. This is because, when constructing the genome module, the conditions were set so that the first eigenvector change was low below a certain threshold.

When the analysis device performs the operations described in Tables 1 and 2 based on the input gene expression data set, genome modules are configured.

The process of constructing the gene network inside the genome module will be described first, and then the process of constructing the network between the modules by connecting the edges between the genome modules will be described.

Genetic network construction

As described above, the genome module is composed of a plurality of genes. Genes in one module form a network that exchanges information. This was called a genetic network.

The genomic module represents a unit of program in the eukaryotic genome. As described above, modules constitute a specific unit in the entire program for living things. Here, the program refers to the process necessary to run the system of living things. Gene networks in the genomic module are elements that connect genes from a specific point of view.

및 모듈에 포함된 임의의 유전자 j의 발현 벡터

를 도시한다. 굵은 실선은 임의의 모듈의 밀도 행렬을 나타낸다. 일반(얇은) 실선은 밀도 행렬에 대한 단위 벡터의 확률 궤적을 나타낸다. 점선은 유전자 i의 배제에 따른 섭동(perturbation)을 나타낸다. 섭동은 역학계에서 주요한 힘의 작용에 의한 운동이 부차적인 힘의 영향으로 인하여 교란되어 일어나는 운동을 의미한다.

는 유전자 j의 정규화된 발현 벡터이므로, 밀도 행렬에 대한 확률은 대응하는 궤적 상에 있다.6 is an example of a density matrix for an arbitrary module in a sample space. 6 is a density matrix of an arbitrary module

And an expression vector of any gene j included in the module

Shows. The thick solid line represents the density matrix of any module. The regular (thin) solid line represents the probability trajectory of the unit vector for the density matrix. The dotted line represents the perturbation following the exclusion of gene i. Perturbation refers to a motion that occurs when a motion caused by the action of a major force in a dynamic system is disturbed by the influence of a secondary force.

Is the normalized expression vector of gene j, so the probability for the density matrix is on the corresponding trajectory.

라고 한다. 유전자 i가 배제되어 모듈이 섭동되면 밀도 행렬은 샘플 공간에서 약간 회전하고, 타원형 모양이 약간 좁아지거나 넓어진다. 밀도 행렬에 있는 다른 유전자 j의 확률은 P_j에서 P_{j ＼i}로 변동된다. 유전자 j가 유전자 i에 강하게 연결된 상태라면, 섭동에 의하여 유전자 j의 확률은 크게 감소하게 된다.Any module is perturbed by the exclusion of gene i. The density matrix of the perturbed modules

It is called. When gene i is excluded and the module is perturbed, the density matrix rotates slightly in the sample space, and the elliptical shape slightly narrows or widens. Probability of another gene j in the density matrix is variation in P _j P _j to _\i. If gene j is strongly linked to gene i, the probability of gene j is greatly reduced by perturbation.

7 is an example showing module perturbation due to the exclusion of gene i in the gene network of any module. Figure 7(A) A shows the genetic network before excluding gene i from any module. B in Fig. 7(B) shows the gene network after excluding gene i. When gene i is removed from the module, gene j linked only to gene i is isolated from the module. Therefore, in the module in which gene i is removed, the decrease in the probability of gene j becomes large. On the other hand, if it is not directly linked to gene i, or even if it is linked to another gene, the decrease in the probability of gene j is small. Therefore, the LOR between the probability of gene j before removal of gene i and the probability after removal of gene i can be used to estimate the link from gene i to gene j.

)를 나타낸다.The odds ratio of probability can quantitatively express the effect between two genes. The log odds ratio (LOR) of the probability is equal to the difference between information contents. Equation 11 below shows the degree of probability variation of the gene j belonging to the same module when gene i is excluded from the module (

).

를 연산한다. 임의의 두 유전자(i와 j)에 대한

가 일정한 임계값을 넘는 경우 유전자 i와 유전자 j는 강한 연결성을 갖는다고 판단한다. 강한 연결성을 갖는 유전자 사이는 에지로 연결한다. 이와 같이 게놈 모듈 내에 존재하는 모든 유전자 사이에서 전술한

를 연산하여 유전자 네트워크를 구성할 수 있다. 예컨대, 도 6은 8개의 조직에 대한 커널 모듈을 도시하는데, 각 모듈에 대한 유전자 네트워크는 전술한 방법을 통해 완성한 것이다.Recall for all possible gene pairs in the genomic module

Computes For any two genes (i and j)

When is over a certain threshold, it is judged that gene i and gene j have a strong connection. Genes with strong connectivity are connected by edges. As described above, among all genes present in the genome module

The gene network can be constructed by calculating. For example, FIG. 6 shows kernel modules for eight tissues, and the gene network for each module is completed through the above-described method.

Table 3 below is a pseudocode for the process of constructing a gene network within a module. Briefly, as described above, LOR is calculated for a pair of genes belonging to an arbitrary module, and an adjacency matrix is generated based on this. The LOR between gene i and all other genes is extracted from the adjacent matrix to calculate the quartile, and the internal threshold th _i of gene i is calculated using the cutoff value. For each gene, the process of giving an edge to a pair of genes having an LOR greater than or equal to the internal threshold value is repeated.

Intermodular network

If the genetic network is a program driven by genes, the program structure of an organism can be represented as a network between modules. As described above, there is an edge between modules in the genomic module network. Here, the edge means that the modules have a certain relationship or connection. Edge can also be viewed as a channel through which a module transfers or exchanges certain information.

로 표현될 수 있다. 여기서

및

는 각각 모듈 i와 모듈 j에 대한 밀도 행렬을 나타낸다. 상대 엔트로피는 항상 음수가 아니고 비가환적이다. 즉,

이다. 상대 엔트로피는 모듈 간 네트워크를 구성하기 위한 정보로 사용된다. 두 개의 밀도 행렬이 동일하다면 상대 엔트로피는 0(zero)이다. 두 개의 밀도 행렬의 차이가 크면 상대 엔트로피 값도 커진다. The process of configuring a network between modules will be described. Relative entropy is measured for all possible pairs of modules extracted from the genetic data set. The description is based on module i and module j. Relative entropy refers to the information gain that module i has for module j. The relative entropy is

It can be expressed as here

And

Denotes the density matrices for module i and module j, respectively. Relative entropy is always non-negative and irreversible. In other words,

to be. Relative entropy is used as information for configuring a network between modules. If the two density matrices are identical, the relative entropy is zero. If the difference between the two density matrices is large, the relative entropy value also increases.

On the other hand, relative entropy is also used in the process of grasping the similarity between modules separated from different organizations. Separated modules from different tissues cannot be directly compared because the sample space is completely different. The relative entropy calculated by mapping a module of one tissue to another tissue represents the difference in density matrix in the same sample space.

If module i has a low information gain with respect to module j, it can be said that module i and module j have a high correlation. In this case, a network is established by connecting the edge between module i and module j.

이다. 여기서,

이다.

> 0이고, i≠ j이다. 아래 표 4는 모듈 간 네트워크를 구축하기 위한 알고리즘에 대한 슈도코드의 예이다.In order to increase the resolution of the relative entropy at a low level, a negative log can be applied to the relative entropy. The relative entropy to which the logarithm is applied is

to be. here,

to be.

> 0 and i≠ j. Table 4 below is an example of pseudocode for an algorithm for building a network between modules.

가 임계값을 초과하지 않는 경우 모듈 i와 모듈 j 사이에 에지를 연결한다. 컷오프(cutoff, C_f)와 관련하여 모듈 간 에지를 결정하기 위한 적절한 임계값을 찾아야 한다. 예컨대, 상기 표 4와 같이

의 제1 사분위수(Q1), 제2 사분위수(Q2) 및 제3 사분위수(Q3)를 사용할 수 있다.For a given module i, a certain threshold is used to determine the association with another counterpart module j. Between module i and module j

Connect the edge between module i and module j if is not exceeding the threshold. In relation to the cutoff (C _f ), we need to find an appropriate threshold to determine the inter-module edge. For example, as in Table 4 above

The first quartile (Q1), the second quartile (Q2), and the third quartile (Q3) of can be used.

The pattern of information exchange between genomic modules can be expressed as a network between modules. The relative entropy between genomic modules in the sample space is measured to confirm the connectivity between modules.

An adjacency matrix composed of relative entropy values not exceeding a threshold calculated using cutoffs and measuring relative entropy from all possible pairs of genomic modules can be prepared. An adjacency matrix can be used to form a network between modules. When constructing a network between modules from an adjacency matrix calculated by reducing the cutoff, the order of module connection depends on the organization type.

It can be said that the modules that are initially connected constitute a seed in each area that constitutes a network between modules. 8 is an example of configuring a module-to-module network with TCGA data sets for eight organizations. Black arrows indicate seeds of kernel domains, white arrows indicate seeds of cell cycle and DNA repair (CCDR) domains. The entropy of the module was expressed in gray scale. The brighter the color, the lower the entropy. In FIG. 8, each node represents one module, and each module is identified by the number of nodes.

For the BRNO and CONO data sets, the seeds of the kernel domain were indicated at cutoff (C _f ) 4.0. The first edge of the BRCA kernel domain did not appear until C _f was 2.2. The inter-module networks of the LUAD, COAD, and READ data sets showed the first edge of the kernel domain at C _f 3.0, 2.8 and 3.0, respectively. For LUSC and OV, the first edge of the kernel domain appeared at C _f 1.9. These results suggest that the network between modules of the tumor may differ from that of normal tissues with respect to the kernel domain.

For the TCGA data set, the inter-module network was reconstructed by experimentally changing C _f . The total number of edges and the number of edges per module were larger in normal tissue than in tumor tissue. This suggests that the tumor's genomic system is simpler than that of normal tissue.

9 is an example of a network between modules of BRNO to which various cutoff values are applied. In FIG. 9, kn is a kernel domain, cc is a CCDR domain, pr is a parenchyma domain, and st is a stroma domain. In FIG. 9, each node has a gray scale value. The brighter the color, the lower the entropy. Lowering the cutoff C _f increases the number of modules with edges and reduces network disconnection. It shows a perfect link between domains before C _f reaches 1.0. In C _f 4.0, the seeds of all domains (kn, cc, pr and st) are already identifiable.

By mapping each module to a gene expression data set of another tissue and exploring the function of basic genes using gene ontology, it is possible to infer a specific biological function of each region divided into a network between modules. BRNO's inter-module network composed of C _f 1.0 clearly explains the relationship between domain domains. The kernel domain kn may control a module pr that performs a real function through the modules m52 and m60. Module m3 relays the information flow between the kernel domain and the CCDR domain.

The function of the epilepsy of the normal breast appears as a module in the st region. In C _f 4.0, the st area is divided into two areas. It can be analyzed that the regions including m38, m64 and m79 are related to adipocytes, and the regions including m27 and m50 perform relay between the interstitial domain and the kernel domain.

The interstitial domain st becomes 6 seeds at C _f 2.5. By analyzing the module, it can be assumed that it performs angiogenesis, immune function (macrophage), extracellular matrix production, and action on adipocytes. In addition, it can be estimated that it also plays a relay role between the kernel domain and the CCDR domain.

C _f Functions between each domain and module can be analyzed through a network between various modules classified according to values.

Several modules located in the central region of the inter-module network connect all domains of the genomic system. It can be regarded as a kind of meta program. This module implies that the extracellular matrix and vasculature are regulated by communication between modules belonging to the epileptic, parenchymal and kernel domains. The genomic system involved in immune function in normal breast tissue appears to be suppressed by other systems.

10 shows an example of mapping the BRNO's inter-module network to modules of other organizations. The color of the node indicates the amount of change in entropy that the BRNO's genomic module has in other tissues. Each node has a grayscale value. Bright colors indicate that the entropy is approximately the same in BRNO and other tissues. Relatively dark colors indicate higher entropy in tissues other than BRNO.

In FIG. 10, (A) is an example of mapping BRNO to CONO, (B) is an example of mapping BRNO to BRCA, (C) is an example of mapping BRNO to LUAD, and (D) is an example of mapping BRNO to LUSC. This is an example of mapping. In FIG. 10 (A) is mapping to other normal tissues, and (B) to (D) are examples of mapping to tumor tissues. The process of mapping the network between modules may be performed based on a genomic module network based on a type of domain and a difference in entropy between modules. In FIG. 10, f denotes an adipose tissue domain.

Referring to FIG. 10(A), in the inter-module network mapped to CONO in BRNO, the kernel domain (kn) shows 0.091-0.182nats entropy. This is similar to the basic module BRNO's entropy (0.017-0.109 nats). However, the CONO kernel domain mapped to BRNO has an entropy of 0.144-0.289 nats, which is slightly higher than the original entropy of CONO (0.016-0.043 nats). This difference may be due to the fact that CONO has a slightly wider kernel domain than BRNO. Thus, it can be assumed that colon tissue is composed of more cell types than breast tissue. In other areas, some of the modules in the parenchyma (pr) and adipose tissue (f) of the normal breast show increased mapping entropy. Thus, it shows that a significant part of the biological program of normal breast tissue should be activated even in the normal colon, but the degree of functional activity may change depending on the input of parameters of the environment and other modules. Except for the kernel domain, modules mapped from CONO to BRNO are less active than standalone modules in CONO. These results suggest that the genomic system of the normal colon is more complex than that of the normal breast.

10(B) to 10(D) show the results of mapping the inter-module networks of BRNO to the tumor data sets BRCA, LUAD, and LUSC, respectively. Although the tumor types were different, the pattern of entropy distribution of the network between the mapped modules was similar in all three cancer data sets.

When all modules included in the parenchymal domain (pr) of BRNO were mapped to each tumor tissue, the entropy was 0.890-1.493 nats. This is much higher than the original entropy of 0.109 nats in BRNO or 0.263 nats of entropy when mapping from BRNO to CONO. Therefore, it can be considered that the parenchymal domain (pr) in the tumor tissue is deformed to the extent that it cannot perform its normal function.

In the case of the CCDR domain, the entropy of the mapped module was 0.754-1.507 nats. Different failure patterns are shown for different tumor types.

In particular, the meta-module linking domains was inactivated in tumor tissue. Meta modules are modules that connect different domains. In the case of mapping from BRNO to LUSC, the entropy for the module m3 was 0.795-1.407 nats, showing the second highest disinitegration after the CCDR domain.

The kernel, CCDR, and parenchymal domains of the genomic system transmit certain information to the stromal domain to control extracellular matrix formation, including angiogenesis (c), immune function (d), and adipose tissue (f) formation. The a and e regions connecting the'interstitial domain (st)' and the'parenchymal (pr), kernel (kn) and CCDR (cc) domain' in the tumor were also very weakened in the tumor tissue. This suggests that the interstitial domains are in a difficult state to communicate with other domains in the tumor tissue. In other words, it means that the epileptic domain is not involved in certain functions by transmitting information to other domains. This is consistent with the abnormal construction of the epilepsy in the tumor tissue without proper control.

From 6 types of tumor tissue gene expression data (BRCA, COAD, READ, LUAD, LUSC, OV), 2 types of normal tissue gene expression data (BRNO, CONO), and 3 types of mixed data (X6CA, X2NO, X6C2N) The resulting genomic module has various levels of entropy. When examined through an experiment, modules with extremely low entropy are separated. The modules with the lowest entropy are (i) the second module in most tissues (m2) and (ii) the first module in breast tumor tissue (BRCA), normal colon tissue (CONO) and ovarian tumor tissue (OV) ( m1). As described above, a genomic module having a lower entropy than other modules corresponds to a kernel module.

11 is an example showing kernel modules for eight organizations. The TCGA data sets for the eight tissues are BRNO, CONO, BRCA, COAD, READ, LUAD, LUSC and OV described above.

As described above, each module contains a plurality of genes. The genes belonging to the module make up the gene network. In FIG. 11, the size of each node is proportional to the number of edges between genes. As can be seen from FIG. 11, although the modules of different tissues, it can be seen that major genes are common. For example, TYR, HBE1, F2, GDF3 and AHSG are commonly included in all tissues, and the remaining genes are also commonly included in more than half of the tissues. In addition, in most tissues, it is possible to confirm the commonality that the gene network is composed mainly of TYR and AHSG. It can be seen that although not completely composed of the same genes in different tissues, the modules (genes constituting the modules) are quite common in different tissues.

The very low entropy of a particular module in all tissues means that it is activated in all cells regardless of phenotype and function. Therefore, the module is meant to perform a common function in all types of cells, and can be considered to be a key component in the genomic system of eukaryotes. This module, which has an extremely low entropy in all tissues and is composed of a common gene, is hereinafter referred to as a kernel module. There may be a plurality of kernel modules, and a set consisting of a plurality of kernel modules is called a kernel domain. The kernel module is presumed to play an important function in the activation of the genomic system by generating gene expression byproducts such as non-coding RNA rather than generating proteins related to a specific protein network.

Experimentally, kernel modules were mapped to different organizations in order to check whether kernel modules are common between different organizations. 12 is an example showing the result of mapping the kernel module of BRNO to another organization. Here, mapping refers to a process of performing the same calculation by extracting data corresponding to a gene belonging to a given module from gene expression data of another tissue.

In FIG. 12, (A) is a basic kernel module of BRNO. (B) is the result of mapping BRNO to CONO. (C) is the result of mapping BRNO to LUSC. (D) is the result of mapping BRNO to BRCA. (E) is the result of mapping BRNO to COAD. (F) is the result of mapping BRNO to READ. (G) is the result of mapping BRNO to X6CA. (H) is the result of mapping BRNO to X2NO. The nodes of the gene network are expressed in grayscale. A node with a light color indicates that the LOR value of the gene is close to 0, and a node with a dark color indicates that the LOR value is negative. The brighter the color, the closer to 0, and the darker the negative value.

When the BRNO kernel is mapped to another tissue, entropy increases if the mapped gene does not exist in the other kernel region or the complexity of the kernel region of the tissue to be mapped is low. Mapping the kernel of BRNO to other tissues means that the necessary calculations are performed by extracting the data of genes contained in the kernel of BRNO from the gene expression data of other tissues. The entropy calculated when the BRNO kernel is mapped to CONO is 0.091 nats. This is much lower than 0.515 nats, the entropy of a randomly selected gene in CONO (normal colon tissue). Therefore, it can be seen that the similarity of kernel modules between two different organizations is quite high. Referring to (B) of FIG. 12, when the kernel of BRNO is mapped to CONO, it can be seen that LOR is close to 0 in most genes. On the other hand, when the BRNO kernel was mapped to tumor data (BRCA, COAD, READ, LUAD, LUSC, OV), the entropy was relatively high, 0.224-0.601 nats. Therefore, the high entropy in the mapping from BRNO (normal breast tissue) to tumor tissue means that the kernel region of the tumor has a distributed direction of its characteristics compared to the normal tissue. Furthermore, similar results were obtained when the kernel modules of CONO were mapped to different data sets.

Domains related to cell cycle and DNA repair (hereinafter referred to as CCDR), which are important in relation to tumors, will be described.

Cell division is an essential process for the development of multicellular organisms from fertilized eggs to somatic cells and for increasing the population of single-celled organisms. Cell division is finely regulated through cell cycle arrest and repair of DNA damage, and dysregulation can lead to abnormal cell growth.

The CCDR domain is made up of multiple modules in normal breast tissue. The 12 modules that make up the CCDR are composed of genes (eg, BUB1) that participate in cell division, and have been shown to be strongly connected to each other through edges. A number of modules composed of these genes were also found in other normal tissues (CONO, X2NO), and a few modules were found in tumor tissues.

When 12 CCDR modules of normal breast tissue are mapped to other normal tissues, the entropy value is generally similar to the original entropy value. In contrast, when the CCDR module is mapped to a tumor data set, the entropy value is raised to the level of random entropy calculated for each data set.

13 shows an example of mapping modules of the CCDR domain of BRNO to modules of other tissues. 13 shows the genetic network for several modules belonging to the CCDR domain. 13 shows m3, m41 and m49 among the modules of the CCDR. 13 is a result of mapping the CCDR module of BRNO to CONO, BRCA and LUSC, respectively. The color of the node is as described in FIG. 12.

13 shows the genetic network within the modules that make up the CCDR domain of a normal breast. The probability of a module contributing to the expression of a gene also indicates how the gene relates to the module. When the CCDR module of the normal breast was mapped to the normal colon (CONO), most of the genes showed a high probability to the module. In contrast, the probability was greatly reduced when the CCDR module was mapped to a cancer data set. The entropy values exceed 1.0 nats when m3, m41 and m49 were mapped to most tumor data sets. This means that the disruption or modification of the CCDR module in the tumor not only leads to cancer cell proliferation, but also prevents it from being in balance with parenchymal and interstitial cellular events beyond the control of the kernel region.

When the results obtained by mapping the CCDR module of normal breast tissue to other normal and tumor tissues are summarized, normal cells are under strict control of the CCDR program, and in tumor tissues, cells in which DNA damage has occurred due to the collapse or modification of the CCDR module. It indicates that the cell cycle continues. When mapping from BRNO to LUAD, the entropy value is more than twice as large as the value when BRNO is mapped to LUSC. This is consistent with the results of previous studies that LUSC has a faster cancer progression rate and a higher probability of mutation than LUAD.

The above-described genomic module network construction process is summarized. 14 is an example of a flow chart 200 for a process of constructing a genome module network. The genome module network constructed in the same manner as in Fig. 14 is referred to as a basic genome module network. The basic genome module network can be constructed in the following process.

The analysis device receives gene expression data (210). Here, the gene expression data may be gene expression data for a normal tissue. The gene expression data is preferably data extracted from a plurality of samples. Gene expression data is data obtained using a technique such as cDNA microarray. Thereafter, the analysis device divides (modulates) genes into specific modules based on the gene expression data (220). This is the process of analyzing gene expression data and dividing genes that make up the genome into specific modules. The analysis device establishes an inter-module network between a plurality of modules (230). In addition, the analysis device builds a gene network between a plurality of genes belonging to the module (240). Gene network construction can be performed after modularization. The analysis device may analyze a network between modules and analyze a genome at a module level (250). The analysis device can analyze the relationship between modules based on the network between modules. In addition, it is possible to analyze the relationship between different samples by mapping networks between modules for different samples. As described above, it can be confirmed that the activity of a specific module or a specific domain in the tumor tissue is weakened compared to the normal tissue.

Furthermore, the analysis device may analyze the genome at the gene level (260). Genetic networks can be used to analyze relationships between genes. Furthermore, it is also possible to analyze gene functions of specific samples by mapping gene networks for different samples. For example, an analysis such as the activity of a specific gene function in a tumor patient, a specific gene inactivity, detection of a gene related to a tumor, etc. may be performed. This can be used to identify genes (markers) related to specific diseases.

In FIG. 14, the order of the inter-module network construction 230 and the gene network construction 240 is not related. Therefore, unlike FIG. 14, the analysis device may first construct a gene network after the modularization 220 of the genome.

15 is an example of a process 300 of calculating an analysis index for sample data using a genome module network. The analysis index includes at least one of the aforementioned SP, MSP, DSP and LOR.

15 shows three DBs. The normal tissue data DB stores gene expression information (first gene expression data described above) for a plurality of normal tissues. The genome module DB stores information generated after constructing a genome module network. The sample data DB stores gene expression information (second gene expression data) to be analyzed. The sample data DB may store gene expression information for tumor tissue. The sample data DB may store gene expression information of a plurality of tumor tissues and individual characteristic information of a corresponding sample. Hereinafter, it is assumed that the sample data DB stores gene expression information of tumor patients. Although FIG. 15 shows three DBs separately, three DBs may be physically located in the same storage device.

^(s))을 계산한다. 또한 상기 유전자 공간에서의 밀도 행렬에 대한 특정 샘플의 확률은 수학식 4를 원용하여 계산한다. 분석장치는 종양 조직 샘플의 유전자 발현 데이터를 획득하고, 이로부터 발현 벡터를 생성할 수 있다. 분석장치는 유전자 발현 데이터로부터 일정한 밀도 행렬을 생성할 수 있다.The analysis device acquires the gene expression vector of the tumor sample from the tumor tissue data DB. The expression vector refers to a one-dimensional array constructed by extracting the expression data of the whole genome or some genes from a specific sample. The analysis device extracts gene expression data of a combination consisting of specific genes from all samples registered in the normal tissue data DB, and uses the above-described Equation 3 to obtain a density matrix in the gene space (

^(s) ) is calculated. Also, the probability of a specific sample for the density matrix in the gene space is calculated using Equation 4. The analysis device may acquire gene expression data of a tumor tissue sample and generate an expression vector therefrom. The analysis device can generate a constant density matrix from gene expression data.

The analysis device acquires first gene expression data for the normal tissue from the normal tissue data DB. The analysis device constructs a genome module network based on the first gene expression data (310). The genome module DB stores information on the constructed genome module network.

The analysis device may extract the index of a specific gene from the genome module DB in order to identify the gene belonging to the target module of the genome module network (320). The genome module DB provides information on which gene a specific module is composed of or information on which module a specific domain is composed. In addition, the genome module DB may provide information on which module or domain the gene is included in, based on the gene. The genome module DB may include a module identifier, a domain identifier, a gene identifier, a table matching a module and a gene, a table matching a domain and a module, a table matching a domain and a gene, and the like.

The analysis device compares and analyzes normal tissues and tumor tissues using information provided by the normal tissue data DB, sample data DB, and genome module DB. The analysis device may generate various indicators to quantify the variation of tumor tissue compared to normal tissue. 15 shows an example in which the analysis device calculates a sample probability (SP), a module sample probability (MSP), a domain sample probability (DSP), and a log odds ratio (LOR).

Gene expression data for a normal tissue using an index in the calculation process may be gene expression data used to construct a genome module network or gene expression data extracted from a separate normal tissue.

)과 샘플 i에서 해당 유전자의 발현 데이터로 구성한 발현 벡터(s_i)를 이용하여 계산한다.The analysis device may calculate the SP (330). SP is a value obtained by quantifying the degree of variation of the genomic system from normal in a sample of an individual cancer patient to be analyzed for all genes included in the whole genome modules. SP represents the degree of variation of the currently input sample based on the whole genome module. That is, SP corresponds to an analysis value for all genes included in the whole genome module. To this end, the analysis device extracts the indexes of all genes included in one or more modules of the whole genome module, obtains a density matrix from normal tissue, and constructs an expression vector with the corresponding gene from specific sample data to calculate SP. SP is expressed with a certain probability for the sample data to be analyzed. The probability of sample i for the corresponding gene set can be expressed as Equation 12 below. This is a density matrix calculated in the gene space defined by the corresponding genes using Equation 3 (

) And the expression vector (s _i ) composed of the expression data of the corresponding gene in sample i.

In fact, the degree of variation of the genomic system of a specific sample i can be determined based on the expression data of genes included in all modules of normal tissue. Therefore, SP can be expressed as in Equation 13 below. That is, SP is equal to P _i calculated in Equation 13.

In Equation 13, G _M refers to the expression matrix of all gene sets included in one or more modules of all modules of a normal tissue, and s _iM refers to an expression vector constructed by identifying the corresponding gene from specific sample data s _i .

In order to calculate the SP, the analysis device identifies all genes belonging to one or more of the genome modules of normal tissues, extracts the expression data of the corresponding gene from the entire sample of the normal tissue reference data DB, and generates a density matrix, a tumor tissue reference sample. SP is calculated by constructing an expression vector by extracting the expression data of the gene from the data.

The analysis device may calculate the MSP (340). MSP means the sample probability for each module. If the above-described SP was a sample probability quantifying the degree of mutation from a normal tissue based on all genes included in the whole genome module, MSP represents the sample probability calculated based on each module. To this end, the analysis device extracts the gene index included in a specific genomic module, obtains a density matrix from a normal tissue, and constructs an expression vector with the corresponding gene from specific sample data to calculate MSP. MSP represents the degree of variability of a particular sample for a particular module of normal tissue. That is, MSP is a value obtained by quantifying the degree of variation of the genomic system for each module in a specific sample. Depending on the disease (specific tumor, etc.), a large variation may appear first in a specific module. Therefore, MSP analysis is also a meaningful index for disease diagnosis or prediction. Further, as will be described later, MSP is also used to classify samples uniformly. MSP can be expressed as Equation 14 below.

에 포함된 유전자 집합의 발현행렬을 의미한다. s_iα는 특정 샘플 데이터 s_i에서 특정 모듈

에 포함된 유전자 발현 벡터를 의미한다. 즉 MSP는 정상 조직의 특정 모듈을 기준으로 확인한 특정 샘플 조직의 게놈 시스템 변이를 나타낸다. 따라서 MSP를 연산하기 위해서는 사전에 게놈 모듈 네트워크가 구축되어야 한다. 유전자들을 포함하는 모듈들이 결정되어야 하기 때문이다.In Equation 14, G _α is a specific module of normal tissue

It means the expression matrix of the gene set included in. s _iα is the specific module in the specific sample data s _i

It means a gene expression vector included in. In other words, MSP represents a genomic system variation of a specific sample tissue identified based on a specific module of a normal tissue. Therefore, in order to calculate MSP, a genome module network must be established in advance. This is because the modules containing the genes have to be determined.

에 속한 모듈에 포함된 유전자 집합의 발현행렬을 의미하고, s_iα는 샘플 데이터 s_i에서 해당 유전자의 데이터를 추출하여 구성한 유전자 발현 벡터를 의미한다.Meanwhile, the analysis device may calculate the DSP (350). A genomic module domain is a collection of genomic modules having similar biological functions and is composed of adjacent modules in a genomic module network. DSP represents a sample probability calculated based on all genes included in one or more modules belonging to a specific domain. To this end, the analysis device extracts the indexes of all genes included in one or more modules belonging to a specific domain, obtains a density matrix from normal tissue data, and constructs an expression vector with the corresponding gene from the sample data to be analyzed to calculate DSP. . DSP represents the degree of variation of a sample to be analyzed for a specific genomic module domain of normal tissue. That is, DSP is a value obtained by quantifying the degree of variation of the genomic system for each domain in the sample to be analyzed. If DSP is described by Equation 14, G _α is a specific domain of normal tissue in Equation 14

It refers to the expression matrix of the gene set included in the module belonging to and s _iα refers to a gene expression vector constructed by extracting the data of the corresponding gene from the sample data s _i .

The analysis device may calculate a log odds ratio (LOR) of a specific gene with respect to the sample probability (360). LOR is a generalized term that means the log ratio of probability according to the presence or absence of a specific condition. The LOR described above refers to the degree of variation of the probability of the genome module of the remaining genes according to the presence or absence of a specific gene in one genome module, and is a value obtained by quantifying the connectivity between genes. On the other hand, fluctuations in sample probability (SP, MSP, DSP) according to the presence or absence of a specific gene in one sample also correspond to LOR. That is, the LOR of a specific gene with respect to the sample probability is a quantification of the effect of the gene on the mutation of the genome system in one sample. LOR is an analysis result of one gene unit. The analysis device can calculate LOR based on several units. (1) LOR _SP is a value obtained by quantifying the degree of influence of a specific gene on the sample probability (SP) for the whole genome module in the sample to be analyzed. (2) LOR _MSP is a quantification of the degree of influence of a specific gene on the sample probability (MSP) for a specific genomic module in the sample to be analyzed. (3) LOR _DSP is a quantification of the degree of influence of a specific gene on the sample probability (DSP) for a plurality of genomic modules belonging to a specific domain in the sample to be analyzed.

15 is an example of constructing a genome module network from normal tissue gene expression data, and then estimating information having mutations in samples compared with normal tissues. Unlike FIG. 15, a genomic module network may be initially constructed from tumor tissue gene expression data, and then samples and tumor tissues may be compared. In this case, it is possible to analyze the sample by determining how different or similar the sample is to the tumor tissue.

Fig. 16 shows an example in which the sample probability (SP) of a plurality of tumor tissue samples is calculated using a genome module of a normal tissue. In Fig. 16, each dot represents the SP of each sample. 16 shows the SP of 248 breast tumor tissue (BRCA) samples. In FIG. 16, A shows the SP of each breast cancer sample for all genes included in the module of normal breast tissue (BRNO). That is, it corresponds to the SP defined in Equation 13 above. In FIG. 16, B represents the probability of each breast cancer sample for all genes, and C represents the probability for all genes not included in any module. As described above, SP represents the degree of variation in the genomic system of the sample. The larger the variation of the genomic system, the lower the SP value, and is located on the left in the graph of FIG. In FIG. 16, when compared with the sample probability for all genes included in the BRNO module, the sample probability for all genes is generally low, but the trend (slope) according to the sample is similar, whereas for all genes not included in any module. The sample probability is about half the slope, so the difference between samples is small. The graph of FIG. 16 shows that genes that are not included in the module in each tumor sample do not reflect mutations in the genomic system well, and it is the genes belonging to the module that lead the mutations in the genome system. Fig. 16 shows that the sample probability, that is, SP, using a gene included in the genome module of a normal tissue, well reflects the variation of the genomic system of each tumor sample.

17 shows an example of comparing tumor tissue sample groups classified based on SP. FIG. 17 is a result of performing a Kaplan-Meier survival analysis using information on the death and death of each patient stored together with gene expression data of 248 breast cancer patients in the tumor tissue reference data DB. FIG. 17(A) is a result of performing a survival analysis by calculating SP in each sample and dividing it into a sample group (A) having an SP of 0.746 or more and a sample group (B) having an SP of less than 0.746. The area around the survival curve represents the 95% confidence interval. FIG. 17(A) shows that the sample group having an SP of 0.746 or more has a significantly higher probability of survival for about 1,700 days (p-value<0.05) than the sample group having an SP of less than 0.746. FIG. 17(B) is a result of performing a survival analysis of each sample group by classifying samples by the presence or absence of expression of an estrogen receptor (ER), which is commonly used as a standard for conventional breast cancer treatment. Fig. 17(B) is a result of performing a survival analysis of a sample group (A) in which ER expression is present and a sample group (B) in which ER is not expressed in breast cancer samples. 17(B) shows that the presence or absence of ER expression is independent of the survival rate of breast cancer patients. 17 shows that by calculating the SP of a sample of a new cancer patient using a genome module of a normal tissue, it is possible to predict the survival of the patient until a specific period.

The analysis device may classify a sample of tumor tissue reference data using the sample probability. 18 is an example of classifying a sample of tumor tissue based on MSP. 18 is an example of classifying samples by calculating the MSP of a tissue sample (BRCA) of 248 breast cancer patients for 85 genomic modules isolated from normal breast tissue (BRNO). 18 is an example of classifying samples through hierarchical clustering of MSPs of breast cancer samples against genomic modules of normal breast tissue. Classification of samples may be performed using all or only part of the MSP for the genomic module of normal tissue or using DSP instead of MSP.

In FIG. 18, a heat map at the bottom shows the result of calculating the MSP for each sample for each module. In FIG. 18, the horizontal axis of the heat map at the bottom corresponds to the breast cancer samples arranged by clustering, and the vertical axis is the classification of genomic modules of normal breast tissue according to domains. In FIG. 18, the upper dendrogram shows the results of clustering of breast cancer samples based on the entire MSP for 85 modules of normal breast tissue. In FIG. 18, the notation starting with R means a sample group according to hierarchical classification. Considering the variation in the number of samples, a total of 8 breast cancer sample groups (R.1.1, R.1.2.1, R.1,2,2, R.2.1, R.2.2.1.1, R.2.2.1.2, R. 2.2.2.1, R.2.2.2.2). In the heat map of FIG. 18, R.2.1 indicates that the sample was composed of a sample having a higher MSP than other sample groups.

The three columns consisting of dots indicated at the bottom of the aqueous phase of FIG. 18 indicate the presence or absence of estrogen receptor (ER), progesterone receptor (PR), and HER2 receptor expression from above. The presence or absence of expression of these receptors is information used in conventional breast cancer diagnosis and drug treatment policies and is a result extracted from sample information provided together with breast cancer gene expression data. Black dots indicate that the receptor is expressed in the sample, and white dots indicate that it does not. In FIG. 18, sample clustering using MSP shows a result that is highly correlated with tumor characteristics and progression. For example, it can be seen that in the sample groups R.1.2.1 and R.1.2.2, triple negative samples in which all ER, PR, and HER2 were not expressed were significantly more than the remaining sample groups. In the case of conventional breast cancer diagnosis, triple negative is classified as the highest risk patient, and conversely, when ER and PR are expressed, it is classified as a relatively low risk patient.

19 is an example of displaying the average MSP for each module of a tumor tissue sample group in a genome module network of a normal tissue. FIG. 19 is a result of displaying the average MSP for each module of the eight breast cancer sample groups classified in FIG. 18 on the genome module network of normal breast tissue. Meanwhile, the genome module network shown in the lower right of FIG. 19 is a result of displaying the average of MSPs for each module of the entire sample of normal breast tissue (BRNO). In FIG. 19, the modules of each genomic module network are displayed in a lighter color as the average MSP is closer to 1 and darker as the average MSP is closer to 0. That is, FIG. 19 shows modules and domains in which mutations in genomic systems mainly occur in each breast cancer sample group. 19 shows that there are differences in domains in which mutations mainly occur according to sample groups. For example, in the sample group R.1.2.2, module variation of the CCDR domain was severely generated. As described above, the CCDR domain is a domain related to the cell cycle and DNA repair. Therefore, if a sample from a specific patient belongs to R.1.2.2, it is expected that drug therapy targeting cell cycle regulation will not be effective. This is because there is a very high possibility that the mechanisms related to the cell cycle are not working properly. In addition, sample group R.2.1 showed similar level of MSP to normal breast tissue in all modules except for module 61. Therefore, patients belonging to R.2.1 are relatively close to normal, and various treatments are expected as early breast cancer patients. In this way, sample classification through MSP and MSP can also be used as a criterion for personalized treatment. Furthermore, it shows that the module and domain classification of the constructed genomic module network is a biologically meaningful technology.

^(S))을 나타내고, 굵은 실선 타원은 해당 유전자 집단에서 유전자 j를 배제하였을 때의 밀도 행렬(

^(S) _＼j)을 나타낸다. 도 20에서 점선은 각 밀도 행렬에 대한 샘플의 확률 궤적을 나타낸다. LOR은 아래의 수학식 15와 같이 표현할 수 있다. 한편, LOR은 특정한 모듈에 속한 특정 유전자에 대하여 연산할 수 있다.FIG. 20 is an example of defining a density matrix of a specific gene group in a gene space and showing the probability of a sample to be analyzed for the gene set. 20 is for explaining values for LOR operation. In FIG. 20, the ellipse of the normal (thin) solid line is the density matrix for the corresponding gene group (

^(S) ), and the bold solid ellipse indicates the density matrix when gene j is excluded from the gene group (

^(S) _\j ). In FIG. 20, a dotted line represents a probability trajectory of a sample for each density matrix. LOR can be expressed as Equation 15 below. On the other hand, LOR can be calculated on a specific gene belonging to a specific module.

LOR is a quantification of the effect of gene j on sample probability (SP, MSP, DSP) in sample data s _i . If the sample probability is a value obtained by quantifying the degree of variation of the genomic system in the sample to be analyzed, the LOR is a value obtained by quantifying the degree of the effect of a specific gene on the sample probability in the sample. In the case of a gene that promotes mutation of the genomic system, the LOR has a negative value, and in the case of a gene that suppresses the mutation, the LOR has a positive value.

21 is an example showing the LOR of a corresponding gene calculated in each sample of a tumor tissue using a density matrix of a specific gene group in a normal tissue. 21 is an example showing the LOR calculated by sequentially restoring and extracting genes belonging to the corresponding gene group from each sample of 248 breast tumor tissues using the density matrix of a specific gene group calculated from gene expression data of normal breast tissue. . In Figure 21, each gene has a different distribution of LOR in 248 breast cancer samples. Fig. 21 shows that the degree of influence of each gene on the mutation of the genomic system varies according to breast cancer patients. In FIG. 21, a breast cancer sample in which the LOR of a specific gene is greatly deviated from 0 means a sample that is highly affected by the gene in the genome system mutation.

에 대한 분석 대상 샘플 s_i의 MSP에 해당하고, P_i＼j는 특정 모듈

에 속한 유전자 중에서 유전자 j를 배제하였을 때 해당 유전자에 대한 샘플 s_i의 MSP를 연산한 값을 의미한다. (3) LOR_DSP경우 P_i는 특정 도메인에 속한 모든 모듈 중 하나 이상의 모듈에 포함된 유전자에 대한 분석 대상 샘플 s_i의 DSP를 연산한 값이고, P_i＼j는 해당 유전자 중에서 유전자 j를 배제하였을 때 샘플 s_i의 DSP를 연산한 값을 의미한다.In Equation 15 (1) In the case of LOR _SP , P _i corresponds to the SP of the sample s _{i to} be analyzed as defined in Equation 13, and P _{i \} j is the corresponding gene when gene j is excluded from the genes belonging to the whole genome module. It means the calculated value of SP of sample s _i for. (2) In the case of LOR _MSP , P _i is a specific module as defined in Equation 14

Corresponds to the MSP of the sample s _i to be analyzed for, and P _{i \j} is a specific module

It means the value obtained by calculating the MSP of the sample s _i for the gene when gene j is excluded from the genes belonging to. (3) In the case of LOR _DSP , P _i is a value obtained by calculating the DSP of the sample s _i to be analyzed for genes included in one or more modules of all modules belonging to a specific domain, and P _{i \} j excludes the gene j from the genes. It means the value calculated by DSP of the sample s _i .

The analysis device calculates the LOR by calculating the sample probability using the density matrix calculated from the expression data of a specific gene combination in normal tissue and the expression vector composed of the corresponding genes in the sample to be analyzed, and calculating the sample probability when a specific gene is excluded. can do. In addition, since it is possible to perform analysis on a gene belonging to a specific module, the analysis device may identify a gene belonging to a specific module by referring to the genome module DB and calculate the LOR for the gene.

In the above description of the LOR, an example of calculating the LOR of each gene of the sample to be analyzed using normal tissue was described. In some cases, it is also possible to calculate the LOR of each gene of the sample to be analyzed using tumor tissue. In this case, the density matrix is obtained from the genome module separated from the gene expression data of the tumor tissue, and the LOR of each gene can be calculated by constructing the expression vector of the sample to be analyzed.

22 is an example of the LOR of a gene calculated in each sample of a tumor tissue using a genome module of a normal tissue. 22 is an example of the LOR of genes calculated in each sample of 248 breast tumor tissues (BRCA) using a normal breast tissue (BRNO) module. The horizontal axis of FIG. 22 corresponds to breast cancer samples arranged according to the result of sample classification using MSP of FIG. 18. That is, FIG. 22 shows the LOR of some genes for each sample of breast cancer in a dot plot. 22 shows an example of the LOR of 20 genes having a constant pattern according to a sample of breast cancer patients. In the graph, all genes marked with × have negative LOR values. In other words, the gene can be said to be a gene that promotes breast cancer. On the other hand, genes marked with ○ have a positive LOR value. Therefore, the gene can be said to be a gene that suppresses breast cancer. As such, the LOR of some genes has a certain tendency even in certain diseases such as tumors. Thus, it can be seen that LOR can be used as a specific indicator for genetic analysis.

On the other hand, the genome module network can be constructed by other approaches other than the basic genome module network described above. Gene expression data can be constantly filtered, and a genome module network can be constructed based on the filtered data. The genomic module network based on the filtered data is referred to as a filtering-based genomic module network.

23 is an example of a flow chart of a method 400 for analyzing sample data based on a filtering-based genome module network.

The analysis device uses two types of gene expression data. One is the first gene expression data for a specific tissue that serves as the basis for constructing a genomic module network. Here, the specific tissue may be a normal tissue or a tumor tissue.

The other is the second gene expression data for the sample tissue to be analyzed. The tissue to be analyzed is a tumor tissue generated at the same location as the above-described normal tissue.

The analysis device constructs the above-described first genomic module network by using the first gene expression data for normal tissue or tumor tissue (410). The first genomic module network is built on the basis of gene expression data for normal or tumor tissue. The first genomic module network includes a module identifier, a gene identifier belonging to a module, connection information between modules, a domain identifier, an identifier of a module constituting a domain, a gene identifier constituting a domain, and connection information between genes belonging to one module (gene network ), etc. When the genome module network is established, the genes belonging to the module are matched, and the connectivity with the module and the connectivity of the genes belonging to the module are completed.

The analysis device performs filtering on the first gene expression data and the second gene expression data of the sample tissue based on a specific module belonging to the first genome module network (420). A detailed filtering process will be described later. The first genomic module network is composed of a plurality of modules, and has connectivity between any one module and at least one module. That is, any one module delivers certain information to at least one module. The filtering process corresponds to a process of blocking (filtering) information transfer to any one module (or a plurality of modules in some cases) that transfers a relatively large amount of information. The specific module from which information transmission is blocked may be the aforementioned kernel module. It has been explained that the kernel module has a lower entropy than other modules and is more likely to be involved in various biological processes. Meanwhile, a plurality of kernel modules may belong to the kernel domain. In this case, filtering may be performed based on at least one of the kernel modules. For example, filtering may be performed based on a module having the lowest entropy among kernel modules.

The specific module to be filtered may be a module having an entropy less than a certain reference value. Here, the reference value may be a different value depending on the type of tissue to be analyzed, the type of disease, and the data collection environment.

The analysis device constructs a second genome module network using the filtered first gene expression data (430). The process of constructing the genome module network is as described above. The second genome module network is built on the basis of the constantly filtered data.

The analysis device performs module mapping on the filtered second gene expression data of the sample tissue based on the constructed second genome module network (440). That is, the analysis device identifies which module each of the second gene expression data of the sample tissue belongs to by using the gene identifier belonging to the specific module in the second genome module network.

The analysis device compares and analyzes the first gene expression data and the second gene expression data based on the module of the constructed second genome module network (450). The analysis device performs analysis based on the entire module of the genome module network, a plurality of modules belonging to the genome module network, or any one module (target module) belonging to the genome module network.

The analysis device compares the difference between the first gene expression data belonging to the target module and the second gene expression data belonging to the same target module. Through this, the analysis device can quantitatively check what kind of change (mutation) has occurred in the sample tissue compared to the normal tissue or tumor tissue. The analysis device may analyze the variation of the sample tissue using the first gene expression data before filtering and the second gene expression data before filtering. In addition, the analysis device may analyze the variation of the sample tissue using the filtered first gene expression data and the filtered second gene expression data.

24 is an example of a process 500 of calculating an analysis index for sample data using a filtering-based genome module network.

24 shows five types of DBs. The first genome data DB stores gene expression information for a plurality of normal tissues or tumor tissues. The first genome data DB stores the above-described first gene expression data. The first genome filtering data DB stores data obtained by uniformly filtering data of the first genome data DB.

The genome module DB stores information generated after constructing a genome module network. The second genome data DB stores gene expression information to be analyzed. The second genome data DB stores the above-described second gene expression data. The second genome data DB may store gene expression information on tumor tissue. The second genome data DB may store gene expression information of a plurality of tumor tissues and individual characteristic information of a corresponding sample. Hereinafter, it is assumed that the second genome data DB stores gene expression information of a tumor patient. The second genome filtering data DB stores data obtained by uniformly filtering data of the second genome data DB.

Although FIG. 24 shows five DBs separately, a plurality of DBs may be physically located in the same storage device.

The analysis device acquires first gene expression data for a normal tissue or a tumor tissue from the first genome data DB. As described above, the analysis device constructs a first genome module network based on the first gene expression data (510).

The analysis device filters the first gene expression data and the second gene expression data based on a specific module belonging to the first genome module network (520). The first genome filtering data DB stores the filtered first gene expression data. The second genome filtering data DB stores the filtered second gene expression data.

The analysis device constructs a new second genome module network based on the filtered first gene expression data (530). The genome module DB stores information on the constructed second genome module network.

The analysis device may extract an index of a specific gene from the genome module DB in order to identify the gene belonging to the target module of the second genome module network (540 ). The genome module DB may include a module identifier, a domain identifier, a gene identifier, a table matching a module and a gene, a table matching a domain and a module, a table matching a domain and a gene, and the like.

The analysis device analyzes individual mutations of a tumor tissue sample based on the second genome module network using information provided by the first genome filtering data DB, the second genome filtering data DB, and the genome module DB. The analysis device may generate various indicators to quantify variation of individual tumor tissues compared to a plurality of normal tissues or tumor tissues. In this process, the analysis device generates an index based on the second genome module network.

Gene expression data using the index in the calculation process may be gene expression data used to construct a genome module network or gene expression data extracted from a separate tissue. In addition, the gene expression data using the index in the calculation process may be filtered gene expression data or pre-filtered gene expression data.

The analysis device may calculate the SP (550). SP is a value obtained by quantifying the degree of variation of the genomic system in a sample of an individual cancer patient to be analyzed for all genes included in the whole genome modules of normal tissue or tumor tissue. SP represents the degree of variation of the currently input sample based on the whole genome module. The analysis device extracts the indexes of all genes included in one or more modules of the whole genome module, obtains a density matrix from normal or tumor tissue, and constructs an expression vector with the corresponding gene from specific sample data to calculate SP. SP is expressed with a certain probability for the sample data to be analyzed. The probability of sample i for the corresponding gene set may be expressed as in Equation 12 described above. However, the analysis device calculates the SP based on the second genome module network. In addition, the analysis device can calculate the SP based on the filtered data.

In order to calculate SP, the analysis device identifies all genes belonging to one or more of the genome modules of the second genome module network, extracts the expression data of the corresponding gene from the entire sample of the first genome filtering data DB, and calculates the density matrix. , SP is calculated by constructing an expression vector by extracting the expression data of the corresponding gene from the second genome filtering data.

The analysis device may calculate the MSP (560). MSP means the sample probability for each module. If the above-described SP was a sample probability quantifying the degree of mutation from a normal tissue based on all genes included in the whole genome module, MSP represents the sample probability calculated based on each module. To this end, the analysis device extracts the gene index included in a specific genomic module, obtains a density matrix from normal or tumor tissue, and constructs an expression vector with the gene from specific sample data to calculate MSP. MSP refers to the degree of variability of a specific sample to a specific module of normal or tumor tissue. That is, MSP is a value obtained by quantifying the degree of variation of the genomic system for each module in a specific sample. Depending on the disease (specific tumor, etc.), a large variation may appear first in a specific module. Therefore, MSP analysis is also a meaningful index for disease diagnosis or prediction. Further, as will be described later, MSP is also used to classify samples uniformly. MSP can be expressed as in Equation 14 above. However, the analysis device calculates the MSP based on the second genome module network. In addition, the analysis device can calculate the MSP based on the filtered data.

에 속한 모듈에 포함된 유전자 집합의 발현행렬을 의미하고, s_iα는 샘플 데이터 s_i에서 해당 유전자의 데이터를 추출하여 구성한 유전자 발현 벡터를 의미한다. 다만 분석장치는 제2 게놈 모듈 네트워크를 기준으로 DSP를 연산한다. 또 분석장치는 필터링된 데이터를 기준으로 DSP를 연산할 수 있다.Meanwhile, the analysis device may calculate the DSP (570). A genomic module domain is a collection of genomic modules having similar biological functions and is composed of adjacent modules in a genomic module network. DSP represents a sample probability calculated based on all genes included in one or more modules belonging to a specific domain. To this end, the analysis device extracts the indexes of all genes included in one or more modules belonging to a specific domain, obtains a density matrix from normal tissue or tumor tissue data, and constructs an expression vector with the corresponding gene from the sample data to be analyzed. Computes DSP represents the degree of variation of a sample to be analyzed for a specific genomic module domain in normal or tumor tissue. That is, DSP is a value obtained by quantifying the degree of variation of the genomic system for each domain in the sample to be analyzed. If DSP is described by Equation 14, G _α in Equation 14 is a specific domain of normal tissue or tumor tissue.

It refers to the expression matrix of the gene set included in the module belonging to and s _iα refers to a gene expression vector constructed by extracting the data of the corresponding gene from the sample data s _i . However, the analysis device computes the DSP based on the second genome module network. Also, the analysis device can calculate the DSP based on the filtered data.

The analysis device may calculate a log odds ratio (LOR) of a specific gene with respect to the sample probability (580). LOR refers to the degree of fluctuation of the probability of the genome module of the remaining genes according to the presence or absence of a specific gene in one genome module, and is a value obtained by quantifying the connectivity between genes. On the other hand, fluctuations in sample probability (SP, MSP, DSP) according to the presence or absence of a specific gene in one sample also correspond to LOR. That is, the LOR of a specific gene with respect to the sample probability is a quantification of the effect of the gene on the mutation of the genome system in one sample. LOR is an analysis result of one gene unit. The analysis device can calculate LOR based on several units. (1) LOR _SP is a value obtained by quantifying the degree of influence of a specific gene on the sample probability (SP) for the whole genome module in the sample to be analyzed. (2) LOR _MSP is a quantification of the degree of influence of a specific gene on the sample probability (MSP) for a specific genomic module in the sample to be analyzed. (3) LOR _DSP is a quantification of the degree of influence of a specific gene on the sample probability (DSP) for a plurality of genomic modules belonging to a specific domain in the sample to be analyzed. However, the analysis device calculates the LOR based on the second genome module network. In addition, the analysis device can calculate the LOR based on the filtered data.

25 is an example of a filtering process 600 for gene expression data. The gene expression data DB stores gene expression data to be filtered. The analysis apparatus may perform filtering on the above-described first gene expression data and second gene expression data, respectively. The analysis device removes a specific component of a linear combination from the gene expression data to be filtered. Various techniques can be used for the removal of certain components of the linear combination. For convenience of explanation, it is assumed that Singular Value Decomposition (SVD) is used.

As shown in Fig. 25, the gene expression data DB holds expression data for n genes for each of m (plural) samples. The analysis device removes a specific component of a linear combination from the gene expression data in the form of a two-dimensional matrix. Singular value decomposition decomposes the m×n matrix A as shown in Equation 16 below. U is a left singular vector matrix, S is a singular value matrix, and V is a right singular vector matrix. Since singular value decomposition is a technique widely known in the field, detailed descriptions are omitted.

The analysis device receives the gene expression data set (610). The analysis device performs SVD on the entire gene expression data set (620).

The analysis device constructs a genome module network using gene expression data of normal tissue (630). The analysis device selects a specific module to be a filtering criterion among modules belonging to the constructed genome module network, and performs SVD calculation for the specific module (640). The analysis device performs SVD calculation on gene expression data belonging to a specific module. For convenience of description, it is assumed that a module serving as a filtering criterion is a kernel module. The analysis device extracts the main vector (column vector) of V (right singular vector matrix) from the SVD decomposition result of the kernel module (650).

The analysis device selects U (left singular vector matrix) and S (singular value matrix) from the SVD results for all gene expression data. Also, the analysis device selects the giveaway vector V ₁ of V (right singular vector matrix) from the SVD result for the kernel module.

The analysis device uses U (left singular vector matrix) and S (singular value matrix) in the SVD results for all gene expression data, and the giveaway vector V ₁ of V (right singular vector matrix) in the SVD results for the kernel module. Filtering is performed (660). Through this, the analysis device prepares a constant filtered gene expression data set (670).

를 생성한다. 경우에 따라서 분석장치는 필터값 벡터를 정규화한다. 최종적으로 분석장치는 전체 유전자 발현 데이터 G의 각 열 데이터에서 필터값 벡터를 감산한 G'를 생성한다. G'가 필터링된 유전자 발현 데이터에 해당한다. 경우에 따라서 분석장치는 G'를 정규화할 수 있다.Table 5 is a number code for the filtering process. Table 5 shows an example of performing filtering based on the first kernel module in the kernel domain. The analysis device multiplies the first eigenvector V ₁ of V (right singular vector matrix) from the SVD result for the entire gene expression data, U (left singular vector matrix), S (singular value matrix), and SVD results for the kernel module. Filter value vector

Create In some cases, the analysis device normalizes the filter value vector. Finally, the analysis device generates G'obtained by subtracting the filter value vector from each row data of the total gene expression data G. G'corresponds to the filtered gene expression data. In some cases, the analyzer can normalize G'.

Hereinafter, an example of a genome module network constructed based on the filtered data will be described. Originally, gene expression data is a value composed of a linear combination. Here, by performing data filtering to remove a specific component, it is possible to find the underlying characteristics that have been obscured by the specific component. A module newly derived from the filtering result is called a base module. It is intended to show that the genomic module network constructed based on the filtered data is also biologically meaningful.

26 is an example of a genome module network generated using data obtained by filtering BRCA. The data obtained by filtering BRCA is called BRX. BRX is the result of filtering BRCA using the main eigenvector of the kernel module. 26 is an example of a genomic module network constructed based on BRX. Hereinafter, a genome module network constructed based on BRX is referred to as a BRX genome module network.

In Fig. 26, modules existing in the genome module network constructed from conventional BRCA (data before filtering) in the BRX genome module network are indicated by circles. In Figure 26, modules present only in the BRX genomic module network are indicated by squares. That is, it can be seen that a base module that was not visible before filtering is derived. As will be described later, the activity of a module related to tumor infiltrating lymphocyte (TIL) among the modules of the genomic module network using BRX was confirmed. For example, module 11 (BRX#11), module 20 (BRX#20), module 39 (BRX#39), and module 52 (BRX#52) of the BRX Genome Module Network It contains a number of chemokine ligands, interleukin receptors involved in the immune response, and interferones. Among the genomic modules, BRX#11 and BRX#20 are genomic modules that exist in the BRCA genome module network before filtering, while BRX#39 and BRX#52 are base modules newly discovered after filtering.

27 and 28 are examples of classifying samples based on MSPs for specific modules of the genome module network of FIG. 26. 27 and 28 show MSP at which the hazard ratio of the survival analysis of the Cox proportional hazard model is maximized when classifying samples based on the MSP for a specific BRX genome module as a reference point. This is an example of classifying into a sample group and a low MSP sample group. 27 is an example of classifying a sample based on the MSP for module 2 (BRX#2) of the BRX genome module network. Fig. 27 is a genetic network of BRX#2 constructed from a high MSP sample group (A) and a low MSP sample group (B) for BRX#2. When the entropy of BRX#2 is calculated from the data before filtering of each sample group, it is confirmed that the entropy of both sample groups is very low. This proves that BRX#2 is a genomic module that also exists in the genomic module network before filtering. Meanwhile, FIG. 28 is an example of classifying samples based on the MSP for module 9 (BRX#9) of the BRX genome module network. Fig. 28 is a genetic network of BRX#9 constructed from a high MSP sample group (A) and a low MSP sample group (B) for BRX#9. When the entropy of BRX#9 was calculated from the BRCA data before filtering of each sample group, it was confirmed that the entropy of the low MSP sample group was more than twice the entropy of the high MSP sample group. This proves that BRX#9 is a basal module not found in the BRCA genomic module search before filtering. 28 shows that the BRX#9 gene network (B) in the low MSP sample group has better connectivity than the gene network (A) in the high MSP sample group.

29 is a survival curve for a specific sample group. 29 is an example of performing Kaplan-Meier survival analysis of breast cancer patient groups classified using BRX#9. In FIG. 29, A denotes a high MSP sample group, B denotes a survival curve and a 99% confidence interval of patients in the low MSP sample group. The biological properties of BRX#9 derived from the BRX genomic module network have not been fully characterized. However, as a result of the Cox proportional probability model survival analysis, the survival rate of patients in the low MSP sample group was 2.4 times higher than that of the patients in the high MSP sample group (p-value <0.05). This is the basis that BRX#9 is a genomic module that allows for meaningful classification of breast cancer samples in relation to breast cancer progression. FIG. 29 implies that a sample with a low MSP for the BRX basal genomic module obtained by filtering BRCA data is a sample with a distance from breast cancer intrinsic characteristics, that is, close to normal tissue. As a result of classifying patients by MSP for several other breast cancer basal modules, a high survival rate was observed in patients in the low MSP sample group.

Sample analysis using kernel module

Describe the gene expression data set used by the researcher. The gene expression data set used is similar to the data used in the experimental process described above. Hereinafter, it will be described again to describe an example of sample analysis using the genome module network. Basically, 8 TCGA gene expression data sets are shown in Table 6 below. Basically, data on 2 normal tissues and 6 tumor tissues were used as follows.

Tissue Naming gene expression data sets normal breast BRNO normal colon CONO breast cancer BRCA colon adenocarcinoma (colon cancer tissue) COAD rectal adenocarcinoma (rectal cancer tissue) READ lung adenocarcinoma (lung adenocarcinoma) LUAD lung squamous cell carcinoma LUSC ovarian cancer OV

Furthermore, a mixed data set synthesized by extracting a plurality of data sets among the eight gene expression data sets was added. The mixed data set is shown in Table 7 below. X2NO is a mixed data set of normal tissue data sets, X6CA is a mixed data set of 6 tumor tissue data sets, and X6C2N is a mixed data set of all eight data sets.

Multiple data sets Mixed data set naming BRNO + CONO X2NO BRCA+COAD+READ+LUAD+LUSC+OV X6CA Mixing all 8 datasets X6C2N

The researcher separated the genomic module using 8 gene expression data sets and 3 mixed data sets (a total of 11 gene expression data sets). The genomic module configuration follows the above-described method of generating a genome module network (in the experiment, a basic genome module network generation technique is used).

Discovery of tumor-specific genomic systems

Genomic module networks were constructed for each of the 11 gene expression data sets. For each of the 11 genomic module networks, modules with low entropy were identified compared to other modules. That is, a kernel module was determined from each of the 11 genome module networks. As described above, the kernel module corresponds to a module having an entropy lower than the reference value. Meanwhile, the reference value for classifying the kernel module may be a value that is adaptively changed according to samples or may be a fixed value. Kernel modules formed around TYR and AHSG were isolated from all 11 gene expression data sets.

In common in the 11 genome module networks, genes included in kernel modules of normal tissues but not included in kernel modules of tumor tissues were identified. Experimentally, all 7 genes were identified. All seven genes were included in the CT antigen (cancer testis antigen) gene. Specifically, the seven genes are MAGEA1, MAGEA3, MAGEA4, MAGEA10, MAGEA12, CSAG1, and CSAG3A. Genes that are strongly presumed to be highly related to tumors based on the kernel module through genomic module network analysis are called CG (critical genes). Genes other than CG in the genes belonging to the kernel module are called CGX. Table 8 below shows the entropy for the kernel module (Kernel), CGX and CG extracted from 11 gene expression data sets.

Except for BRCA, OV and X6C2N, the entropy of the kernel module in the remaining data sets were all 0.1 bits or less.

는 유전자 발현 데이터 세트

와

의 관계이다.

은 m개의 샘플에서 개별 샘플의 특성에 관련된 요소이다. 그러므로,

은 2ⁿ 전사 상태들에 대하여 개별 샘플의 측정 과정의 편차를 포함한 상이한 특성을 모두 포함하게 된다.

을 특이값 분해(SVD)로 인수분해하면

이다. 여기서, 고유값인

는 대각 행렬로 포함된 샘플이 동질(homogeneous)이 아니라면, 단위 행렬(identity matrix) I가 될 수 없다. 따라서, 유전자 발현데이터로 계산한 폰노이만 엔트로피

는 전사 상태의 엔트로피

보다 크게 된다. 이는 두 가지 이유로 발생할 수 있다. 첫째, 동질적 샘플 군에 대해 작위적 샘플 선택이나, 샘플 취급 과정에 문제가 있는 경우

의 편중에 의해 발생한다. 둘째, 타입이 다른 샘플이 혼합된다면 단일 T라는 가정에서 벗어나게 된다.Transcriptional state matrix

Is the gene expression data set

Wow

Is the relationship.

Is a factor related to the properties of individual samples in m samples. therefore,

Will contain all of the different properties, including variations in the measurement process of individual samples for 2 ⁿ transfer states.

Factoring with singular value decomposition (SVD)

to be. Here, the eigenvalue

If the samples included in the diagonal matrix are not homogeneous, then cannot be the identity matrix I. Therefore, von Neumann entropy calculated from gene expression data

Is the entropy of the warrior state

Becomes larger. This can happen for two reasons. First, if there is a problem with the random sample selection or sample handling process for a homogeneous sample group

It is caused by the bias of Second, if samples of different types are mixed, the assumption of a single T is deviated.

In the case of X6C2N, since the data is a mixture of normal and tumor tissues, the heterogeneity of T is large and the entropy of the kernel module increases. Based on such theories and examples, it is natural that the entropy of the kernel is high in BRCA and OV, which are estimated to be mixed with various subtypes.

Meanwhile, since the kernel module extracted from all tumor tissues does not contain CG, the entropy of the kernel module and CGX is bound to be the same. Meanwhile, in BRNO, CONO and X2NO, the entropy of the kernel module and CGX was almost the same. The entropy of normal tissue in kernel module and CGX was lower than that of tumor manipulation, but the difference was relatively small. On the other hand, the entropy of CG shows a difference of at least 10 times larger (the entropy of the tumor tissue is very high) between normal and tumor tissues. It can be interpreted that in the tumor tissue, CG is separated from the kernel module and in severe cases (eg, LUSC) is completely disrupted.

Comparing the kernel module, CGX and CG in BRNO, CONO, and X2NO, the entropy of the kernel module and CGX in X2NO, which is the mixed data of BRNO and CONO, increased significantly from 1.7 to 2.0 times. The increase in entropy of kernel modules is due to CGX, which occupies a significant portion of kernel modules. As mentioned above, this is an increase in entropy as samples having different transcription states T are mixed. Therefore, the CGX of BRNO and CONO imply that they are in different transcriptional states and are the starting point of the difference in the biological properties of the two tissues.

On the other hand, the entropy of CG of X2NO was insignificant compared to that of BRNO and CONO. The fact that the increase in entropy of CG was minimal despite mixing samples of BRNO and CONO of different types suggests that CG is a functional genomic system before the biological differences between the two tissues are created. Comparing entropy with normal tissue, it can be seen that the increase in entropy of CG is significantly greater than that of CGX. This suggests that the development of tumor tissue is due to the disruption of CG.

Relative entropy and angular divergence (AD) were measured to measure the degree of CG deviation from the tumor tissue. The measured relative entropy and each divergence are shown in Table 9 below. In tumor tissue, in order to calculate the relative entropy of CGX and CG under the same conditions as normal tissue, CG was added to the gene set of the kernel module isolated from the tumor tissue to construct the kernel of the tumor tissue.

는 상대 엔트로피

를 나타내고,

는

로서,

는 A의 밀도 행렬의 제1 고유 벡터를 나타낸다.In Table 9, kr represents a kernel module,

Is the relative entropy

Represents,

Is

as,

Denotes the first eigenvector of the density matrix of A.

및 CG의 상대 엔트로피

를 연산하였데, 모두 종양 조직에서 증가하였다. 또한, CGX에 대한 CG의 상대 엔트로피

도 같은 정도로 종양 조직에서 증가하였다. The relative entropy of CGX for the kernel module of the tumor tissue and the kernel module of the normal tissue generated in this way

And relative entropy of CG

Was calculated, all increased in tumor tissue. Also, the relative entropy of CG to CGX

The same degree increased in tumor tissue.

와 CG의 밀도 행렬의 제1 고유 벡터

사이의 각도 발산(AD_CG,kr),Also, the first eigenvector of the density matrix of the kernel module

And the first eigenvector of the density matrix of CG

Divergence of the angle between (AD _CG,kr ),

와 CGX의 밀도 행렬의 제1 고유 벡터

사이의 각도 발산(AD_CGX,kr) 및 CG의 밀도 행렬의 제1 고유 벡터

와 CGX의 밀도 행렬의 제1 고유 벡터

사이의 각도 발산 (AD_CG,CGX)을 측정하였다. CG의 유전자 수가 CGX에 비해 작아 종양 조직의 커널 모듈에 추가된 CG의 영향은 비교적 제한적일 것이다. First eigenvector of density matrix of kernel module

And the first eigenvector of the density matrix of CGX

The angle divergence between (AD _CGX,kr ) and the first eigenvector of the density matrix of CG

And the first eigenvector of the density matrix of CGX

The angular divergence between (AD _CG,CGX ) was measured. The number of genes in CG is smaller than that of CGX, so the effect of CG added to the kernel module of tumor tissue will be relatively limited.

In fact, AD _CGX,kr in tumor tissue was smaller than AD _CG,kr . In other words, AD _CG,kr was significantly greater in tumor tissue than in normal tissue (anova test: p = 0.001). In addition, AD _{CG and CGX} were also significantly greater in tumor tissue (anova test: p = 0.001). These results indicate that CGX and CG are closely related in normal tissues, whereas in tumor tissues, CGX and CG are separated and CG has a tendency to collapse.

에 대한 조직 B의 샘플 군의 밀도 행렬

의 상대 엔트로피

를 계산하였다. 조직 간 CGX의 상대 엔트로피를 계산한 결과는 아래 표 10과 같다. In order to determine the degree and direction in which the gene sets CGX and CG are transformed into a genomic system in normal and tumor tissues, the relative entropy between the tissues of CGX and CG was calculated. First, the density matrix of the sample group of tissue A in the gene space composed of the genes of CGX of tissue A to calculate the relative entropy between tissues of CGX.

The density matrix of the sample group of tissue B for

Relative entropy of

Was calculated. The results of calculating the relative entropy of CGX between tissues are shown in Table 10 below.

Excluding the mixed data set, the tissue with the smallest relative entropy for CGX in BRNO (normal breast tissue) is BRCA (breast cancer), and the tissue with the smallest relative entropy for CGX in CONO (normal breast tissue) is COAD (colon cancer). And READ (rectal cancer). In addition, the tissue with the smallest relative entropy for CGX of LUAD (lung adenocarcinoma) is LUSC (lung squamous cell carcinoma), another type of lung cancer. This trend appears in the dendrogram generated by the relative entropy between tissues for CGX.

30 is a result of clustering based on the relative entropy between eight tissues for CGX. FIG. 30 is a diagram of an aqueous phase generated based on the relative entropy between eight tissues for CGX. Fig. 30 is not a division between a tumor tissue and a normal tissue, but a tumor tissue and a normal tissue as an origin are grouped together to accurately classify according to the type of tissue.

The results of calculating the relative entropy of CG between tissues are shown in Table 11 below.

Excluding the mixed data set, the tissue with the smallest relative entropy for CG in BRNO (normal breast tissue) is another normal tissue, CONO (normal colon tissue), and the tissue with the smallest relative entropy for CG in BRCA (breast cancer) These are LUAD (lung adenocarcinoma), COAD (colon cancer) and READ (rectal cancer). In addition, LUSC (pulmonary squamous cell carcinoma) and OV (ovarian cancer), which tend to be malignant, had the lowest relative entropy for CG. This trend appears in the water level generated by the relative entropy between tissues for CG.

31 shows the results of clustering based on the relative entropy of 8 tissues for CG. FIG. 31 is a diagram of an aqueous phase generated based on the relative entropy between eight tissues for CG. Fig. 31 accurately distinguishes the normal tissue and the tumor tissue, and the tumor tissue is classified according to the characteristics of the tumor.

Tumor tissue results in overall phenotypic variation in tissues and in all cells, including tumor cells. Attempts have been made to explain such mutations as changes or mutations in the degree of expression of specific genes or gene groups. However, conventional studies cannot clearly distinguish between normal and tumor tissues based on a single factor.

On the contrary, referring to FIGS. 30 and 31, analysis of the kernel module (measurement of CG and CGX mutations) after the construction of the genome module network was able to classify not only normal tissues and tumor tissues, but also the types of tissues. The functional location of the kernel module within the genomic system and the results of the study suggest that tumors begin with CG departure and CGX mutation.

Hereinafter, whether the analysis based on the kernel module (CG and CGX) is meaningful, an additional analysis will be described.

1. Hormone receptor related to breast cancer

The integrity of the genomic system that CG and CGX make up in tumor tissue is associated with malignant transformation of cells. Thus, CG and CGX may have an overall effect on the normal function of cells. Researchers analyzed CG and CGX in breast cancer to verify whether these predictions were correct.

In general, the treatment method of breast cancer is determined using the expression of hormone receptors. In the case of hormone receptor negative breast cancer, hormone therapy treatment is not effective, and hormone therapy drugs are known to be effective in the case of hormone receptor positive breast cancer. However, in order to apply these treatments collectively to all breast cancer patients, treatment effects are often different for each individual.

The researchers analyzed the relationship between the integrity of CG and CGX and the expression of hormone receptors in breast cancer. In the analysis below, the researcher classified groups based on positive and negative expression of receptors in BRCA and BRNO samples related to breast cancer, and analyzed based on the classified groups.

CG and CGX ( n = 44 each) were extracted for each positive and negative sample group for ER (estrogen receptor), PR (progesterone receptor), and HER2 (human epidermal growth factor receptor 2). Entropy was calculated. The entropy of CG was measured to be 0.1522, 0.1399 and 0.1986 bits in the positive group of ER, PR and HER2, respectively, and 0.4666, 0.3781 and 0.2604 bits in the negative group of ER, PR and HER2, respectively. That is, the entropy of CG was measured lower in the positive group than in the negative group in each hormone receptor. The entropy of CGX was measured to be 0.2378, 0.2172 and 0.2708 bits in the positive group of ER, PR and HER2, respectively, and 0.2269, 0.2775 and 0.2417 bits in the negative group of ER, PR and HER2, respectively. That is, the difference between the positive and negative groups in the entropy of CGX was insignificant.

In addition, BRCA is a tumor tissue, and CG is separated from CGX, and the relationship between the degree of separation and the expression state of each receptor was estimated as the relative entropy between CGX and CG. Looking at the relative entropy in BRNO, the relative entropy of CG to CGX was 0.025 bits (n = 28) and that of CGX to CG was 0.062 bits ( n = 28).

Among the BRCA samples (n = 44), in ER+, PR+, and HER2+ (positive group), the relative entropy of CG to CGX was significantly higher, 1.4845, 1.5023 and 3.5093 bits, respectively. However, in the voice sample group, it increased to 7.1480, 5.4585 and 3.0739 bits. CGX for CG was 1.5616, 1.5307 and 4.0471 bits in the positive sample group of the three receptors, and 6.4663, 5.2620 and 3.1832 bits in the negative sample group. Compared to the positive sample group, ER and PR significantly increased in the negative sample group, but decreased in HER2.

The relative entropy for BRNO of CG and CGX was calculated in the positive or negative sample group of each receptor of BRCA (Table 12 below).

In general, the relative entropy of BRCA to BRNO of the genomic module refers to the degree of deviation from the steady state. Therefore, for ER and PR, when the degree of departure from the normal state of CG was large, the receptor was not expressed, but the difference in the degree of departure of CGX with respect to the expression of the receptor was not large. In HER2, when CG escaped from the normal state, the expression of the receptor was suppressed, but in CGX, the departure from the normal state rather promoted the expression of the receptor. In other words, the expression of HER2 increases when CG leaves small and CGX leaves large. On the other hand, ER and PR were expressed when both CG and CGX had a small departure from BRNO.

Meanwhile, the MSP of each sample can be calculated based on CG and CGX. MSP _CG and MSP _CGX refer to the MSP of each sample calculated based on CG and CGX, respectively. FIG. 32 is a result of calculating MSP _CG and MSP _CGX in 248 BRCA samples and showing the distribution of MSP _CG and MSP _CGX according to the expression of the receptor. FIG. 32(A) is an example showing the distribution of MSP _CG according to receptor expression in a BRCA sample, and FIG. 32(B) is an example showing the distribution of MSP _CGX according to receptor expression in a BRCA sample.

In Figure 32(A), the distribution of MSP _CG shows a significant difference in each of the positive and negative sample groups of three receptors, and the median value of MSPCG in the positive sample group of each receptor is compared to the median value of MSP _{CG in} the negative sample group. It was significantly larger. In FIG. 32(B), MSP _CGX was significantly greater in ER+ and PR+ than in each negative sample group. On the other hand, in the case of HER2, although not significant, MSP _CGX was higher in the negative sample group than in the positive sample group.

Meanwhile, in the case of the TNBC (triple negative breast cancer, ER-/PR-/HER2-) sample group, MSPCG and MSPCGX showed a significant difference from the other sample groups (p-value <0.005).

To further confirm this, the BRCA samples were divided according to the levels of MSP _CG and MSP _CGX , and a significant difference in receptor expression was verified by a binomial test. When the MSP _CG was less than 0.9932, only the expression of ER was significantly reduced compared to the whole BRCA. When MSP _CGX was greater than 0.9885, the expression of ER and PR was significantly increased compared to the whole BRCA. In the case of HER2, there was no significant change according to the levels of MSP _CG and MSP _CGX . In normal breast tissue, CG and CGX are contained in kernel modules. Therefore, even in breast cancer, the two genomic systems must have a close connection.

Researchers gave the BRCA sample group 'hh', 'hl', 'lh' and 'll' in accordance with the levels of MSP and the MSP _{CGX CG} (Table 13 below).

In Table 13, Th _CG and Th _CGX refer to each reference point for dividing a sample using MSP _CG and MSP _CGX . Hereinafter, it is assumed that Th _CG = 0.9932 and Th _CGX = 0.9885.

Of the four BRCA sample groups, lh was excluded because the number of samples was only 11. After calculating the expression frequencies of ER, PR, and HER2 in each of the remaining three BRCA sample groups, the significance between the positive and negative groups of each receptor was verified by a binomial assay (Table 14 below).

As a result of the binomial test between the negative and positive groups of ER and PR, as expected, in the sample group ll , that is, the BRCA sample group with low MSP _CG and low MSP _CGX , the frequency of negative samples of both receptors was significantly higher as expected.

On the other hand, the frequency of negative HER2 was significantly higher in sample group ll , and the frequency of positive was significantly higher in sample group hl , that is, high MSP _CG and low MSP _CGX . This result is consistent with that in Table 12, the relative entropy of CG for BRNO in the HER2-positive sample group is low and that of CGX is high compared to the HER2-negative sample group.

The expression of ER and PR was dependent on the degree of disruption of the genomic systems of CG and CGX, and in particular, the dependence of CG on the genomic system was large. The relative entropy for CG and CGX of BRNO in BRCA sample group hl where the expression of HER2 is significantly increased are 0.0023 and 0.4018 bits, respectively. In other words, in the HER2+ sample, the degree of disruption of CGX related to cell differentiation is large, but the integrity of CG, which is the starting point of tumorigenicity, is relatively maintained. can do.

On the other hand, in BRCA sample group ll , the relative entropy of BRNO for CG and CGX was increased to 0.6986 and 0.4357 bits, respectively, indicating that the genomic system capable of expressing HER2 was disrupted. Thus, a decrease in HER2 expression may appear in sample group ll . On the other hand, in the BRCA sample group hh, that is, the sample group that maintains the characteristics of differentiated epithelial cells well because the genomic systems of both CG and CGX were well preserved, many HER2- samples were included compared to HER2+ ( Table 14). That is, the HER2- sample appears in both sample groups hh and ll.

Therefore, the malignancy of breast cancer associated with the expression of HER2 was high in the HER2+ sample and in the sample group ll of the HER2- sample. On the other hand, since it is highly likely that the expression of ER and PR in sample group ll is lowered, breast cancer patients with high malignancy have a high probability that all three receptors will eventually become triple negative.

Taking these results together, the genomic system of CG and CGX is related to the expression of ER, PR and HER2 by being involved in the degree of differentiation and tumorigenicity of cells in breast cancer.

2. CG and CGX Associated genomic module

TCGA's normal breast tissue gene expression data (BRNO) consisted of 28 samples. The tissues determined to be pathologically normal from the tissues extracted by surgery for breast cancer were used. Researchers calculated MSPs for CG and CGX for BRNO samples. MSP _CG and MSP _CGX were distributed in 0.9911 to 0.9989 and 0.9921 to 0.9984, respectively. The lower values of the two probability distributions are less than the maximum values for breast cancer, implying the potential for transformation into a tumor in the genomic system, even if pathologically normal.

Therefore, MSP _CG and MSP _CGX were used for hierarchical clustering of BRNO samples. 33 is a result of clustering BRNO samples based on MSP _CG and MSP _CGX . 33 is an aqueous phase diagram of a BRNO sample generated based on MSP _CG and MSP _CGX of a BRNO sample. The aqueous phase of the BRNO sample suggests that the genomic systems of the BRNO sample may have heterogeneous properties from each other.

FIG. 34 is an example showing the relative entropy of each genomic module with respect to the kernel module in the BRNO inter-module network. 34 is the relative entropy of each genomic module with respect to the kernel module calculated from the data of the entire BRNO sample. The domains and modules belonging to the domains in the BRNO inter-module network of FIG. 34 are as follows (Table 15).

BRNO's genomic system is linked around the kernel module. meta means the above-described meta domain. In some modules of the adipose tissue formation domain (adipo), the relative entropy was increased to 0.9265 ~ 4.3553 bits, and in the modules included in the epithelial domain (epi), the relative entropy was increased to 1.0170 ~ 1.9703 bits.

In order to confirm that the modules of the epithelial domain (epi) are deviated from the dominance of the kernel module, the sample of the branch (R.2.2.1.2) having values close to 1 in the MSP _CG and MSP _CGX of the BRNO sample of FIG. Relative entropy for the kernel module of each module was calculated by using (sample indexes 15, 11, 18, 8, 24, 7, 21, 10 and 19).

FIG. 35 is an example of displaying the relative entropy of each genomic module with respect to a kernel module based on a part of a sample in the BRNO inter-module network. 35 is a relative entropy of each genomic module for a kernel module calculated from samples belonging to R.2.2.1.2 among the branches of the BRNO sample water degree. As shown in FIG. 35, the relative entropy (0.0496 ~ 0.2131 bits) of the modules belonging to the epithelial domain (epi) was significantly reduced, and the relative entropy of some modules belonging to the adipoforming domain (adipo) was also reduced. This result suggests that the kernel module dominates most modules of the genome system in the case of samples whose genomic system is estimated to be the most normal among normal breast samples.

Using some of the BRNO samples, the researcher calculated the relative entropy of the BRNO modules for CGX, which constitutes a large part of the kernel module. FIG. 36 is an example of displaying the relative entropy of each genomic module to CGX of a kernel module based on a part of a sample in the BRNO inter-module network. FIG. 36 is a relative entropy of each genomic module to CGX of a kernel module calculated from samples belonging to R.2.2.1.2 among the branches of the BRNO sample award degree. Referring to FIG. 36, it can be seen that the relative entropy of each genomic module with respect to the entire kernel module of FIG. 35 is substantially similar.

Using some of the BRNO samples, the researcher calculated the relative entropy of the BRNO modules for the CG that constitutes a large part of the kernel module. 37 is an example showing the relative entropy of each genomic module with respect to the CG of the kernel module based on a part of the sample in the BRNO inter-module network. 37 is the relative entropy of each genomic module to the CG of the kernel module calculated from samples belonging to R.2.2.1.2 among the branches of the BRNO sample water degree. In all modules, the relative entropy for CG was greater than the entropy for CGX. Therefore, it can be assumed that the CG is indirectly connected to the BRNO modules.

가 0.0196 bits로 제일 컸다. Among the branches of the BRNO sample water degree of FIG. 33, that is, R.1, R.2.1, R.2.2.1.1, R.2.2.1.2, and R.2.2.2, R.1 has only two samples, It is not suitable for calculating entropy. Therefore, as a result of calculating the relative entropy of CG to CGX in each branch except R.1, the relative entropy of CG to CGX in R.2.1

Was the largest at 0.0196 bits.

의 평균치는 0.2175, 0.3582, 0.0843 및 0.1094 bits이다. R.2.2.1.1과 R.2.1에서 상대적으로 높은 값을 갖는다. 표준편차는 각각 순서대로 0.2943, 0.6219, 0.1485 및 0.1739이다. 즉, R.2.2.1.1 에서 유난히 큰데, 이는 상피 형성 도메인(epi) 및 지방 형성 도메인(adipo)의 모듈들의 상대 엔트로피가 집중적으로 커지는데 기인한다. 따라서, R.2.1에서는 CG의 이탈로 인해 각 게놈 모듈에 대한 CGX의 지배력이 전반적으로 감소된 것으로 추정되는 반면, R.2.2.1.1에서는 보다 복잡한 기작이 개입되었을 것으로 추정된다.The relative entropy of each genomic module to the kernel module in each sample branch R.2.1, R.2.2.1.1, R.2.2.1.2 and R.2.2.2 of the BRNO sample award degree

The average values of are 0.2175, 0.3582, 0.0843 and 0.1094 bits. It has a relatively high value in R.2.2.1.1 and R.2.1. The standard deviations are 0.2943, 0.6219, 0.1485 and 0.1739, respectively. That is, it is exceptionally large in R.2.2.1.1, which is due to the intensive increase in the relative entropy of the modules of the epithelial domain (epi) and the adipose domain (adipo). Therefore, in R.2.1, it is estimated that the dominance of CGX for each genomic module is generally reduced due to the departure of CG, whereas in R.2.2.1.1, a more complex mechanism is estimated to be involved.

3. Analysis of predictive tumor occurrence for normal samples

In order to more clearly confirm the characteristics of BRNO samples, samples of BRNO were combined with samples of breast cancer (BRCA) to calculate MSPs for CG and CGX of BRNO, and hierarchical clustering was performed. 38 shows the results of clustering based on the MSP of the BRNO sample and the BRCA sample calculated for CG and CGX of BRNO. The lower part of FIG. 38 is an enlarged view of the names of samples belonging to the classified branches.

Most of the BRNO samples (19) were separated from the BRCA samples, but among the branches of the BRNO sample aqueous phase, samples 4, 5 and 6 included in R.2.2.2 were mixed with the samples of BRCA in the immediately adjacent branch. Samples 12, 20, and 26 of R.2.1 were separated into more distant branches and mixed with samples of BRCA. Samples 14 and 23 of R.1 and sample 16 of R.2.1 were separated into considerably distant branches and mixed with samples of BRCA. These results suggest that mutations may proceed in the genomic system before tumor cells can be identified pathologically. In other words, samples of BRNO that are classified in the same branch as those of BRCA can be seen as samples in which mutations into tumor tissues have begun (possible tumor development), unlike normal tissues.

In order to confirm the variation of the genomic system in the BRNO sample, MSPs of samples of BRNO and BRCA were calculated for all BRNO genome modules and a level plot was created. 39 is an example of a level plot clustered based on MSP of samples of BRNO and BRCA calculated for all BRNO genome modules. In FIG. 39, the horizontal axis is the hierarchical clustering result of samples of BRNO and BRCA, and the vertical axis is the hierarchical clustering result of all BRNO genome modules. As expected, samples of BRNO and BRCA were separated into different branches in the aqueous phase except for some. In BRNO samples 2 and 3, which deviated from the branch consisting of the BRNO sample, the MSP for the modules of the epithelial domain (epi) was reduced to the MSP level of the BRCA samples, and in the BRNO sample 9, the MSP for modules 76 and 81 was 0.3463. And 0.3118, but the MSP for the remaining modules was also between 0.8 and 0.9, which is difficult to recognize as normal. In addition, the MSPs of BRNO samples 20, 15 and 22 indicate that in these samples, modules 61 related to fat formation were collapsed, and modules 49 and 33 were partially collapsed. BRNO sample 12 is also included in the same branch as the BRCA sample, away from the branch composed of the BRNO sample. The MSP of BRNO sample 12 showed the lowest value in 30 of all 85 BRNO genomic modules, which is the most among BRNO samples.

These 30 modules mainly belong to domains related to the cell cycle, stroma and angiogenesis. FIG. 40 is an example showing a module having the lowest MSP among the MSPs of BRNO sample 12 for all BRNO genome modules in a network between BRNO modules. 40 is an example of a module having a low MSP of BRNO sample 12 in an inter-module network constructed using all of BRNO samples.

These results suggest that TCGA's BRNO samples are unlikely to be all in the normal range in terms of variations in the genomic system. In particular, samples 2, 3 and 9 show that the modules of the epithelial domain (epi) are mutated. In general, an increase in relative entropy can occur through collapse and mutation of the module. The decay can be distinguished by an increase in entropy and the variance by the angular divergence of the eigenvector. The results of calculating the angle divergence of each sample for each module (34, 52, 53, 70, 76, 81, 84) of the epithelial domain (epi) are shown in Table 16 below.

Although the angular divergence is large only in 2, 3 and 9 of the samples of R.2.2.1.1 of the BRNO sample water degree, the angular divergence was calculated for all the samples included therein for the calculation of entropy. Entropy was increased in R.2.2.1.1. Therefore, it can be seen that in the samples of R.2.2.1.1, mutation and disruption of the epithelial domain (epi) occur simultaneously.

As described above, the connectivity between genomic modules can be estimated by relative entropy. Density matrices of the two genomic modules are calculated from gene expression data obtained from a group of samples having the same properties for a specific aspect, and relative entropy is measured from these density matrices. Since the relative entropy is obtained in the sample space, it makes it possible to estimate the connectivity in the entire sample. However, the relative entropy cannot estimate the connectivity between modules within individual samples. In order to estimate the connectivity between modules in individual samples, the researcher developed an indicator called SSMC (Single Sample Modular Connectivity). SSMC will be described prior to the following description.

는 각 유전자 세트의 샘플 벡터

,

및

로 구성된다. 각 모듈의 밀도 행렬을 각각 ρ_c,a와 ρ_c,b로 정의하고, 두 게놈 모듈의 통합 밀도 행렬을 ρ로 정의했을 때, 각 모듈의 엔트로피 사이의 관계는

이다. 이때 두 모듈 사이에 상대 엔트로피가 충분히 낮다면

이다. 그 반대의 경우에는

로 근접한다. 따라서 전자의 경우 ρ는 단축(minor axis)의 고유 값에 대한 장축(major axis)의 고유값의 비가 큰 타원이고, 후자의 경우 상대 엔트로피가 증가함에 따라 비율이 감소하여 점차 원형에 가까워진다. 따라서 두 게놈 모듈의 통합 밀도 행렬 ρ에 대한 샘플 i의 확률 p_i는

이고， 모든 샘플의 확률은 모듈 a 및 b에 대한 확률이 높건 낮건 감소하게 된다. 한편，두 모듈이 합쳐진 것에 대한 샘플 i의 확률 p_i는 유전자 세트 a, b 및 공유되는 유전자 세트 c의 함수로 아래의 수학식 17과 같이 표시된다.In order to estimate the connectivity between genomic modules in individual samples, the investigator assumed a set of genes c shared by the two modules and a set of genes a and b unique to each module. Sample vector of sample i

Is the sample vector for each gene set

,

And

Consists of When the density matrix of each module is defined as ρ _c,a and ρ _c,b , and the integrated density matrix of the two genomic modules is defined as ρ, the relationship between the entropy of each module is

to be. At this time, if the relative entropy between the two modules is low enough

to be. Vice versa

Close to Therefore, in the former case, ρ is an ellipse in which the ratio of the eigenvalue of the major axis to the eigenvalue of the minor axis is large, and in the latter case, the ratio decreases as the relative entropy increases and gradually approaches the circle. Therefore, the probability p _i of sample i for the integrated density matrix ρ of the two genomic modules is

And the probability of all samples is module a And the probability for b decreases, whether high or low. Meanwhile, the probability p _i of sample i for the two modules combined is the gene set a, It is expressed as Equation 17 below as a function of b and shared gene set c.

Here, the integrated density matrix ρ of the two genome modules is the density matrix ( ρ _c , ρ _a and ρ _b ) of each gene set and the asymmetric matrix generated between them ( ρ _ca , ρ _cb , ρ _ac ) as shown in Equation 18 below. , ρ _ab and ρ _bc ).

, here,

,

,

,

이다.

to be.

Therefore, p _i can be expressed as Equation 19 below.

,

,

이며,

및

를 제외한 나머지 항은 각 모듈 개별에서 샘플의 특성을 규정한다. 따라서 샘플 i에서 두 모듈 사이의 연결성과는 관계가 없다. 두 모듈에 의해서 공유되는 유전자 세트 c는 두 모듈 사이에 연결성을 증가시키며 특히 γ가 클수록 연결성은 커진다.

는 두 모듈에 의해 공유되지 않은 유전자들에 의한 연결성을 표시하며 공유되는 유전자의 수가 적을 때 중요한 의미를 갖는다. 따라서 두 모듈을 합친 것에 대한 샘플 i의 확률 p_i는 두 모듈의 무결성과 연결성을 종합적으로 표현한다. p_i가 SSMC 지표에 해당한다.here,

,

,

Is,

And

The remaining terms except for stipulate the characteristics of the sample in each module. Therefore, in sample i, the connectivity between the two modules is irrelevant. The gene set c shared by the two modules increases the connectivity between the two modules, especially the greater γ, the greater the connectivity.

Indicates connectivity by genes that are not shared by the two modules, and has an important meaning when the number of shared genes is small. Hence, the probability p _i of sample i for combining the two modules together expresses the integrity and connectivity of the two modules. p _i corresponds to the SSMC indicator.

SSMC was calculated in each BRNO sample to examine the relationship between the kernel module module and the modules of the epithelial domain (epi). Table 14 below shows the SSMC between BRNO module 6 and the modules of the epithelial domain (epi) in each sample of BRNO. Table 15 below shows the SSMC between the BRNO module 68 and the modules of the epithelial domain (epi) in each sample of BRNO.

In Tables 17 and 18, it was confirmed that the BRNO samples 2, 3, and 9 belonging to the BRNO sample water level R.2.2.1.1 had reduced SSMC in all modules of the epithelial domain (epi). This means that the exchange of information for the module has been cut off or reduced. The modules closest to the epithelial domain (epi) on the network between BRNO modules are 6 and 68.

In BRNO samples 2, 3 and 9, modules 6 and 68 had reduced modules of the epithelial domain (epi) and SSMC. As a result of calculating SSMC between modules 6 and 68 and the kernel module, a slight decrease in connectivity was observed in Sample 3, but no decrease in connectivity was observed in Samples 2 and 9. Therefore, it is suggested that another module affected by the kernel module is mediated between modules 6 and 68 and the epithelial domain (epi). As a result of examining the relationship between several nearby modules and kernel modules, and between 6 and 68, module 18 was the most prominent.

The SSMC between the kernel module and module 18 had an average of 0.9811 ± 0.0050 excluding samples 2, 3 and 9, and decreased to 0.9488, 0.9328 and 0.9460 in samples 2, 3 and 9, respectively. As a result of calculating SSMC for CG, which is a set of genes constituting the kernel module, samples 2, 3, and 9 were 0.8796, 0.8374, and 0.8762, and the average of the rest excluding them was 0.9661 ± 0.0087. As a result of calculating SSMC for CGX, which is a set of genes constituting the kernel module, samples 2, 3 and 9 were 0.9397, 0.9205, and 0.9362, and the average of the rest excluding them was 0.9789 ± 0.0051. These results indicate that the mutation of CG in the kernel module damaged the connectivity between the kernel module and module 18 and blocked the connection between modules 6 and 68 and the epithelial domain (epi). As a result, the functional properties of the epithelial cells were lost, resulting in the loss of epithelial cells. It means that it can be transformed into other forms.

, i=21,30,39,81)로부터 각각의 제1 고유 벡터 (

)를 계산하였다. 각 BN2211 모듈의 유전자 공간에서 BRNO 샘플 j (j = 1, ... ， 28)의 발현 벡터

와

사이의 각도를 측정한 결과는 아래 표 19와 같다.As a result of newly extracting modules from BRNO samples of R.2.2.1.1 (hereinafter referred to as BN2211) containing samples 2, 3 and 9, all 87 modules were obtained. In order to confirm the biological function of the modules obtained from BN2211, the modules of BRNO were mapped to BN2211 and the relative entropy with the BN2211 module was calculated. The BN2211 module refers to a module belonging to an inter-module network built using samples belonging to BN2211. 41 is an example of a level plot clustered based on the relative entropy of BN2211 modules for BRNO modules mapped from BRNO to BN2211. Modules (34, 52, 53, 70, 76, 81 and 84) of the epithelial domain (epi) of BRNO had a low relative entropy only in module 87 of BN2211. That is, in BN2211, it was confirmed that most of the modules of the epithelial domain (epi) were collapsed. On the other hand, modules 21, 30, 39 and 81 of BN2211 show that there is no matching module in BRNO. 42 is the distribution of MSP of BRNO samples for modules 21, 30, 39 and 81 of BN2211. As shown in FIG. 42, in the BN2211 modules 21 and 81, the MSP of Sample 9 was small, but in other cases, the MSP was relatively high. Thus, modules 21, 30, 39 and 81 are relatively well maintained in integrity in BRNO samples 2, 3 and 9. Meanwhile, as a result of calculating the SSMC of each BRNO module and the BN2211 modules in each BRNO sample, the connectivity was reduced in samples 2, 3, and 9. In particular, in Samples 2 and 3, the isolation of these modules was remarkable, but in Sample 9, the connectivity between the kernel module and the CCDR domain (SSMC: 0.03-0.88) was not sufficient, but the isolation was relaxed. In these samples, all modules of BN 2211 except for modules 21, 30, 39, and 81 had sufficient connectivity with kernel modules, and had high connectivity with most other modules except for the epithelial domain (epi). Density matrices of these modules in all BRNO samples to see if the biological functions of modules 21, 39, 30 and 81 of BN2211 are transformed (

, i=21,30,39,81) from each first eigenvector (

) Was calculated. Expression vector of BRNO sample j (j = 1, ..., 28) in the gene space of each BN2211 module

Wow

The results of measuring the angle between are shown in Table 19 below.

It can be seen that the angle of the expression vector in BRNO samples 2, 3 and 9 has a significantly different value from that of other samples, so that the direction is significantly different. This shows that in these samples of BRNO, the biological properties were transformed due to a change in the orientation, not the size (degree) of the phenotype generated by the genomic modules.

Modules 21, 30, 39 and 81 of BN2211 are free from direct or indirect domination by kernel modules. BNRF is a genetic data set created by removing the influence of the kernel module in BRNO. The BNRF is a result of filtering BRNO using the main eigenvector of the kernel module using the above-described filtering technique. Therefore, modules that are not affected by the kernel module in BN2211 should be detected in the BNRF. The modules of BN2211 were mapped to the BNRF, and the modules of the BNRF and the relative entropy were measured. The modules of BNRF refer to modules belonging to the inter-module network constructed using the genetic data set of BNRF. 43 is an example of a level plot clustered based on the relative entropy of BNRF modules for BN2211 modules mapped from BN2211 to BNRF. The modules of BNRF with low relative entropy for modules 21, 30 and 39 of BN2211 were 8, 27, 45 and 13 (see Table 20 below).

For module 81 of BN2211, module 17 of BNRF showed the lowest relative entropy at 0.139 bits. The distribution of the angles of sample vectors of BRNO relative to the first eigenvector of the density matrix in the gene space of mapping these modules of BNRF to BRNO is similar to BN2211 modules 21, 30, 39 and 81 in BRNO samples 2, 3 and 9 It showed a marked difference from other samples. These results suggest that these modules of BN2211 and BNRF are directly related to cellular transformation.

Genes constituting modules 21, 30, 39 and 81 of BN2211 were investigated to elucidate the transformation of the biological properties of cells (see Table 21 below).

In order to analyze the extent to which BN2211 contributes to the variance of modules 21, 30, 39 and 81 in R.2.2.1.1, including BRNO samples 2, 3 and 9, R. For 2.2.1.2, the LOR of genes included in each module was calculated and compared. 44 is a result of calculating LOR for genes of modules 21, 30, 39, and 81 of BN2211. 44 shows the significance of BRNO sample groups R.2.2.1.1 and R.2.2.1.2 based on the genes of the module of BN2211. 44(A) is the LOR of module 21 of BN2211, FIG. 44(B) is the LOR of module 30 of BN2211, FIG. 44(C) is the LOR of module 39 of BN2211, and FIG. 44(D) is of module 81 of BN2211 Represents LOR. A number of genes showing statistically significant differences were included, and most of them were genes closely related to EMT (Epithelial Mesenchymal Transition). Modules 21, 30, 39 and 81 of BN2211 are designated as'EMT modules'.

In order to trace the role of the EMT module on breast cancer, a level plot of the relative entropy of the modules of BN2211 versus the modules of BRCA was created. FIG. 45 is an example of a level plot in which BRCA modules mapped from BRCA to BN2211 are clustered based on the relative entropy of modules of BN2211. As shown in FIG. 45, there was no BRCA module that could correspond to modules 21, 30, 39, and 81 of BN2211. The relative entropy of the BRCA modules for the mapped BN2211 modules was calculated. 46 is an example of a level plot clustered based on the relative entropy of BRCA modules for BN2211 modules mapped from BN2211 to BRCA. Even in this case, it was not possible to find modules of BRCA that can correspond to modules 21, 30, 39, and 81 of BN2211. Therefore, these modules can be seen to be activated only in rare samples, which is difficult to detect in BRCA, or activated only in some cells in one sample.

To determine this, the MSP of the BRCA samples for these modules of BN2211 was calculated and the sample award degree was created. 47 is a result of clustering based on the MSP of BRCA samples for some modules of BN2211. 47 is a result of clustering based on the MSP of BRCA samples for modules 21, 30, 39 and 81 of BN2211. As shown in FIG. 47, BRCA samples belonging to R.2.2 have a relatively high MSP. Therefore, it is reasonable to interpret that the frequency of cells in which these modules are activated in one sample is not low, but that the sample in which these modules are activated is difficult to detect for limited reasons. These results are consistent with the notion that BRCA is an aggregate of various subtypes of breast cancer.

In order to detect the occurrence of EMT in BRCA, samples of BRCA were grouped above and below thresholds (ThCG = 0.9976 and ThCGX = 0.9724) for MSP _CG and MSP _CGX , respectively (4 groups), MSP _CG and MSP _CGX. When all were higher than the threshold, BAHH, MSP _CG was high, but MSP _CGX was low, BAHL, MSP _CG was low and MSP _CGX was high, BALH, and all low, BALL. The threshold value was a value that maximizes the hazard ratio of the survival analysis of the Cox proportional hazards model in MSP _CG and MSP _CGX, respectively. Fig. 48 is a survival curve for a group in which samples of BRCA are classified based on MSP _CG and MSP _CGX . 48 is an example of performing a survival analysis between the sample group BAHL (MSP _CG >Th _CG and MSP _CGX <Th _CGX ) and the remaining samples. In the sample group BAHL, the probability of survival was most rapidly decreased. In the samples of BAHL, the risk ratio was significantly higher at 3.628 (p = 0002), and most of the deaths died before 1,000 days. Compared to BALH, the probability of dying within 3,000 days increased by 18.38 times ( p = 0.0004).

In the BRNO sample, the relative entropy of CG to CGX was 0.0251 bits, but increased to 0.0782 bits in BAHL, which is very low compared to the relative entropy of CG to CGX in the entire BRCA sample as 4.6542 bits as shown in Table 9. to be. In contrast, the relative entropy of CGX to CG was significantly increased in BRNO and BAHL to 0.0616 bits and 1.5938 bits, respectively. The kernel modules independently separated from BAHL and separated are genes AHSG, APOA2, APOC3, APOH, ASGR2, C14orf1 15, CSAG1, CSAG3A, DCT, DPPA4, F2, GATA1, GDF3, HBE1, HBG1, HEMGN, IGFBP1, LIN28 , MAGEA1, MAGEA10, MAGEA12, MAGEA3, MAGEA4, PASD1, PRODH2, RHAG, RHOXF2B, SERPINA7, SILV, TM4SF5 and TYR. Except for CG, the rest of the genes have some different composition from BRNO's CGX, and the relative entropy of CG for them is very low, 0.0167 bits. Therefore, unlike general tumor tissues in which CGX and CG are separated from the kernel module and CG is collapsed, in BAHL, CG is well preserved and not separated from CGX. Instead, CGX is mutated to enhance connectivity with CG. It can be seen that mutation has occurred.

To elucidate the relationship between the mutation of the kernel module and the phenotype of BRCA, the relationship with the N stage, which means the degree of lymph node metastasis among breast cancer stages, was examined in four sample groups (Table 22 below).

It can be seen that only BAHL has a significant decrease in N0 without lymph node metastasis. Therefore, this result means that an increase in MSP _CG and a decrease in MSP _CGX causes an increase in tumor cell motility, which may cause EMT as in the case of BRNO.

The researcher extracted genomic modules from BAHL samples to identify the genomic system that induces EMT in tumor cells. All 63 genome modules were extracted, and the information content of the activated genomic system was 0.7312 bits, compared to 1.0150 bits of BRCA. It can be seen that the degree of mixing of breast cancer subtypes decreased. However, larger than 0.3712 bits of BRNO, it can be seen that there is a mixture of subtypes or collapse of the genomic system. The mapping entropy of the BRNO modules to BAHL was 0.9760 ± 0.5821 bits, and the mapping entropy of the BAHL modules to BRNO was 0.4234 ± 0.3038 bits. These results imply that the modules of BRNO in BAHL are collapsed, and it can be estimated that the modules of BAHL are activated according to the degree in BRNO. In order to confirm the heterogeneity of BAHL with respect to BRNO, the relative entropy of BAHL modules with respect to the modules of BRNO was calculated. 49 is an example of a level plot clustered based on the relative entropy of BAHL modules for modules of BRNO mapped from BRNO to BAHL. The CCDR domain of BRNO corresponds to module 46 light 57 in BAHL (area A), and kernel module domains (area B) and interstitial domains (area C) are also clearly distinguished. On the other hand, the epithelial domain (epi) is widely distributed in BAHL with a relatively high relative entropy without clear distinction (region D). In BAHL, it means that epithelial cells have transformed into a tumor state.

The investigator calculated the MSPs of BRCA samples for the modules of BAHL to elucidate the relationship between the modules isolated from BAHL and lymph node invasion of tumor tissue cells. 50 is an example showing the MSP distribution of BRCA samples for the BAHL module. The BRCA samples were divided into N0 group and N1 or higher group of N stage, and the significance of the difference in MSP distribution in each module was verified by the Kruskal Wallis test. 50 shows that there is a significant difference between the MSP distribution of the N0 group and the MSP distribution of the N1 or higher group in BAHL modules 26, 32, 43, 53 and 60. The distribution of MSP of BRNO samples for these modules is shown in FIG. 51. 51 is an example showing the MSP of BRNO samples for BAHL modules 26, 32, 43, 53 and 60. As shown in Fig. 51, the MSP of these BRNO samples in the BAHL module 60 is markedly lowered.

In the BAHL kernel module, CGX excluding CG is mutated differently from BRNO's CGX, so module 60 in the BAHL sample is controlled by the mutated CGX, has a high MSP, and node invasion can be accelerated. On the other hand, in the BRNO sample with the CGX in a steady state, a problem occurs when the module 60 is disconnected from the information exchange with the CGX. In fact, the SSMC of BAHL's module 60 mapped to CGX and BRNO in BRNO had an average of 0.9374 ± 0.0117 excluding samples 2, 3 and 9, while the average of the rest of the samples was reduced to 0.8124 ± 0.0195. In conclusion, in samples 2, 3 and 9 of BRNO, BAHL module 60 induces cell transformation by deviating from regulation due to interruption of information exchange with normal CGX, whereas in BRCA, mutated CGX modulates BAHL module 60 to transform cells. Cause.

Since the N0 group and the N1 or higher group of BRCA have relatively different genomic systems, the two sample groups also have differences in the distribution of the LOR of the constituent genes of each module (Table 23 below). PIM2 of module 26 induces EMT in breast cancer cells by continuously activating STAT3. NT5E (CD73) is also closely related to the activity of EMT.

The relationship between the N-stage and MSP was elucidated by constructing the award map with the MSPs of the BRCA samples for these modules of BAHL. 52 shows the results of clustering based on the MSP of BRCA samples for BAHL modules 26, 32, 43, 53 and 60. In FIG. 52, BRCA samples with N stage of N0 are concentrated in R.2.2, and in R.2.1.2.2.2.2.2.2.1 (red box), the ratio of N0 to the whole (35.4%, p = 00427) Is small, and the ratio of N2 (26.6%, p = 0.0122) is large, indicating that the fluctuation of these modules is involved in the lymph node invasion of tumor tissue cells. In addition, 22 of the 28 BAHL samples were included here, suggesting that kernel module mutations (MSP _CG rise and MSP _CGX fall) are related to mutations in these modules.

4. DNA damage in breast cancer

Mutations in genes that are most likely to be responsible for oncogenesis generally occur extensively in tumors. In the BRCA data of TCGA, various mutations occurred in varying degrees depending on the sample. Researchers tracked whether kernel module mutations were related to the occurrence of mutations, and also explored related genomic systems.

The causes of mutations are truly diverse and depend on the internal and external environment of the tissue. Therefore, the occurrence of mutations inevitably varies depending on individual samples. An average of 155 mutations occurred in one sample of BRCA, with a standard deviation of 181.5. Mutation is presumed to occur due to a combination of several factors, so finding the main factors is difficult. As described above, since the kernel module mutation starts with the separation of CG and CGX, the relationship between the MSP and the mutation of the BRCA sample for BRNO CG and CGX can be elucidated.

First, looking at the distribution of the incidence of mutations with MSP _CG as a threshold value, if MSP _CG is greater than the median, the mutation is 188.1 ± 229.0, and if it is small, it is significant as 121.6 ± 107.2 (Cruscal Wallis test, p = 0.0006; anova test. , p = 00037). In the case of MSP _CGX , it was 188.7 ± 232.9 when it was greater than the median, and 121.0 ± 98.0 when it was smaller than the median value (Cruscal Wallis test p = 0.0005; anova test, p = 00031). The mutation of the kernel module is the starting point of the tumor tissue, showing that the MSP of CG and CGX in each sample is associated with the mutation. In addition, as a result of comparing the frequency of mutations when the SSMC showing the connectivity between CG and CGX in each sample was greater than the median value and when it was less than the median, the incidence of mutations was as low as 115.6 ± 94.2 when the SSMC was large. Was as large as 194.1 ± 232.8. In spite of the standard deviation keoteum and this difference is statistically significant (Kruskal Wallis test, p = 1.0 x 10- 5; anova test, p = 00006) was.

In order to more precisely analyze the relationship between the incidence of mutations and mutations in the kernel module, the investigator made a water degree of the samples with MSPCG and MSPCGX of BRCA, and classified the samples into 10 groups. 53 is a result of clustering BRCA samples according to MSP _CG and MSP _CGX .

The relationship between the median incidence of mutations in each sample group and the relative entropy, which indicates the degree to which CG and CGX of BRCA deviate from BRNO were looked at. Figure 54 shows the results of a linear regression analysis between the median incidence of mutations in the BRCA sample group and the relative entropy of CG and CGX of BRCA to CG and CGX of BRNO. Fig. 54(A) shows the results of a linear regression analysis between the median incidence of mutations in the BRCA sample group of Fig. 53 and the relative entropy of CG of BRCA to CG of BRNO. Fig. 54(B) shows the results of a linear regression analysis between the median incidence of mutations in the BRCA sample group of Fig. 53 and the relative entropy of CGX of BRCA to CGX of BRNO. In the linear regression of the mutation frequency of the BRCA sample with respect to the relative entropy of CG, r ² was 0.87, but in the case of CGX, it was as small as 0.10. In the case of dual linear regression of mutation incidence for the relative entropy of CG and CGX, r ² increased to 0.93 (p = 8.0 x 10-5). Therefore, among the mutations in the kernel module, CG departure is mainly related to the occurrence of mutations, but the mutations in CGX also have some influence.

를 계산하였으며，BRCA 샘플 그룹 j의 돌연변이 빈도의 중앙값 m_j를 계산하였다. 모듈 i에서 m_j를 S_ij에 대해 선형 회귀 분석을 시행하고,

를 BRNO의 모듈 간 네트워크에 표시하였다. 도 55는 BRNO의 모듈 간 네트워크의 각 모듈에 BRCA 샘플 그룹의 돌연변이 빈도에 대한 선형 회귀 분석 결과를 BRNO의 모듈 간 네트워크에 도시한 예이다. 도 55는 BRNO의 각 모듈에서 도 53의 BRCA 샘플 그룹의 돌연변이 빈도에 대한 선형 회귀 분석을 시행하고 각 모듈의 r²를 BRNO의 모듈 간 네트워크에 도시한 예이다. 돌연변이의 빈도와 커널 모듈인 모듈 2는 CG와 CGX에서 본 바와 같이 밀접한 연관(r² = 0.90, p = 3×10^-5)을 보여주고 있다. 커널 모듈 도메인에 속하지 않는 모듈 51은 r²이 0.89로 두 번째로 컸다(p = 4.3×10^-5).Considering the constituent genes of the kernel module, it is not possible to find the reason for mutations caused by CG departure itself. Therefore, it would be reasonable to search for mutations in the modules in which CG and CGX mutations can cause mutations. To search for a module related to mutation, it is reasonable to compare the degree of deviation from the steady state and the frequency of mutation. Therefore, the relative entropy of BRCA to the module of BRNO for the sample group j of BRCA.

Was calculated, and the median value m _j of the mutation frequency of the BRCA sample group _j was calculated. _Perform a linear regression analysis on S _ij for m _j in module i,

Is shown in BRNO's inter-module network. FIG. 55 is an example of a linear regression analysis result of the mutation frequency of a BRCA sample group in each module of the BRNO inter-module network in the BRNO inter-module network. FIG. 55 is an example of performing a linear regression analysis on the mutation frequency of the BRCA sample group of FIG. 53 in each module of BRNO, and showing r ² of each module in the inter-module network of BRNO. The frequency of mutations and module 2, the kernel module, show a close correlation (r ² = 0.90, p = 3×10 ^-5 ) as seen in CG and CGX. Module 51, which does not belong to the kernel module domain, has an r ² of 0.89, the second largest (p = 4.3 × 10 ^-5 ).

In order to further explore the modules related to mutation, multiple linear regression analysis was performed by including modules other than module 51. When modules 44 and 58 were included in module 51, the r ² and p-values were 0.99 and 2.4×10 ^−6, respectively. In order to estimate the influence of CG and CGX on these three modules, linear regression analysis was performed on the relative entropy of these modules for BRNO in BRCA sample groups of FIG. 52 and the relative entropy of CG and CGX, and r ² was calculated ( Table 24 below).

Relative entropy, which is the degree to which module 51, which is most associated with mutations, deviates from normal, has 92% dependence on the linear combination of the degree to which CG and CGX deviate. On the other hand, modules 58 and 44 are 63% and 54%, respectively, and r ² for the frequency of mutations are 0.70 and 0.44, respectively, and the p-values are significant as 0.003 and 0.036, respectively.

를 계산하였다. 또한 BRNO의 샘플 공간에서의 확률(

)을 계산하고，BRCA의 샘플 공간 j에서 유전자 i의 승산비(odds ratio)

를 얻었다. BRCA의 샘플 공간 j 에서

에 대해 돌연변이 빈도의 중앙값 m_j의 선형 회귀 분석을 하였다. BRNO 모듈 51 에서 유의한( p값 < 005) 유전자는 20개 (MAGEA12, MAGEA1, CAD, MKRN3, PROC, ASFIB, RELL2, C21orf125, PFKFB4, SPAG4, C9orf100, C8G, MCM7, E2F1, ORC1L, NY-SAR-48, DLL3, SERPINF2, MAGEA4 및 DUSP9)였으며，r²가 가장 큰 것은 MAGEA12(0.93)이고 가장 작은 것은 DUSP9(0.40)이었다. 모듈 58에서는 18개의 유전자의 승산비에 대해 돌연변이의 빈도가 유의한 선형관계를 보였으며， NME5의 r²가 0.57로 가장 컸다. To search for genes with a high degree of association with mutations, the probability of genes in the sample space was calculated. Separation of 10 sample spaces according to the aqueous _phase of the samples generated by BRCA's MSP _CG and MSP _CGX , and the probability of gene i in sample space j

Was calculated. Also, the probability (

), and the odds ratio of gene i in the sample space j of BRCA

Got it. In BRCA's sample space j

For, a linear regression analysis of the median mutation frequency m _j was performed. In BRNO module 51, there were 20 ( p value <005) genes (MAGEA12, MAGEA1, CAD, MKRN3, PROC, ASFIB, RELL2, C21orf125, PFKFB4, SPAG4, C9orf100, C8G, MCM7, E2F1, ORC1L, NY-SAR) -48, DLL3, SERPINF2, MAGEA4 and DUSP9), with the largest r ² being MAGEA12 (0.93) and the smallest being DUSP9 (0.40). In Module 58, the frequency of mutations showed a significant linear relationship with the odds ratio of 18 genes, and the r ² of NME5 was the largest at 0.57.

In the genomic system, it is necessary to separate and identify the characteristics of parts related to DNA damage response (DDR) and repair. The genome system related to DDR may have a wide range of regions, and BRCA samples were classified into four groups according to the frequency of mutations, and the relative entropy of each sample group for the BRNO module was calculated. Figure 56 shows the relative entropy for the BRNO module of the four BRCA sample groups sorted according to mutation frequency. Section A is the CCDR domain of BRNO, which contains genes involved in the repair of DNA damage.

However, only in the modules of section A and module 3, the relative entropy of the sample group with a mutation frequency of 500 or more was the second largest, but in other modules, the relative entropy of this sample group was the highest. The relative entropy of modules 51, 58 and 44 as seen above has a linear relationship with the frequency of mutations, and r ² is close to 1, making it suitable for DDR, and the function of the constituent genes is far from DNA damage recovery. Taking these results together, the frequency of mutation is determined by the integrity of the genome system operating DDR and the integrity of the genome system operating the DNA damage recovery system.If the frequency of mutations is 500 or more, the integrity of the DNA damage recovery system While the is relatively good, the collapse of the operating system of DDR is intensifying. Relative entropy is the highest in all modules except the CCDR domain. Therefore, the incidence of mutations is dependent on the DDR and DNA damage repair system. In this study, a linear combination was attempted between the frequency of mutations and one of the modules 51 and the CCDR domain. The linear combination of module 43 of the CCDR domain and module 51 corresponding to the DDR system had an r ² of 0.93 and a p-value of 8.0 × 10 ^-5 . Fig. 57 is a result of prior regression analysis of mutation frequency and relative entropy of BRCA for BRNO modules 43 and 51. R.1, which has the highest mutation frequency in the BRCA sample group divided into MSP _CG and MSP _CGX , has a very high dependence on module 51, unlike other sample groups, as shown in FIG. 57. Therefore, when the frequency of mutation is extremely high, it is due to a mutation in the DDR system rather than a DNA damage recovery system.

As previously suggested, mutations in kernel modules cause mutations in the connected genomic system, resulting in DNA mutations. It is of great significance to clarify the mechanisms that the most sensitive module 51 and module 43 of the DNA damage repair system are related to the kernel module among DDRs widely distributed in the genomic system. Researchers have elucidated the relationship between CG and CGX and the relative entropy of modules 43 and 51 in BRCA sample groups (Table 25 below).

Module 43 is dependent on the relative entropy between CG and CGX in the sample space of BRCA, but module 51 is also dependent on these relative entropy, but is mainly dependent on the relative entropy of BRCA to BRNO's CG. In the end, different aspects of kernel module mutations have different effects on the collapse of DDR and DNA damage repair systems, leading to the frequency of redundant mutations.

The analysis process using the above-described genome module network and kernel module of the genome module network is summarized. 58 is an example of a process 700 of analyzing a kernel module of a genome module network. 58 is a process of determining a kernel module using gene expression data of a normal tissue and gene expression data of a tumor tissue, and generating a constant index based on differences of genes belonging to the kernel module.

The analysis device constructs a first genome module network using a gene expression data set for normal tissue (711). The analysis device determines a kernel module (first kernel module) of the first genome module network (712). In addition, the analysis device constructs a second genome module network using the gene expression data set for the tumor tissue (721). The analysis device determines a kernel module (second kernel module) of the second genome module network (722).

As described above, the kernel module is a module having a lower entropy than other modules in the inter-module network. The analysis device may determine a module having an entropy lower than a certain reference value as a kernel module. In this case, an appropriate value may be used according to the type of tissue, the type of tumor, and the characteristics of the gene expression data set. In addition, the analysis device may calculate entropy for all modules of the inter-module network, and may determine a kernel module based on a lower module having a difference between the upper modules and a predetermined value. However, since the kernel module corresponds to a module having a significantly lower entropy than other modules, it is preferable to determine a clearly distinguished group as a kernel module.

The analysis device includes a first gene group (CG) that is present in the first kernel module (normal tissue) but does not exist in the second kernel module (tumor tissue), and the remaining second gene group (CGX) excluding CG from the kernel module. Determine (730).

Thereafter, the analysis device may determine various mutation indices based on the first gene group (CG) and the second gene group (CGX) (740). The mutation index can determine the relative entropy between CG and CGX, MSP for CG and CGX, etc. Furthermore, the analysis device may determine the connectivity (SSMC) between the kernel module (CG and/or CGX) and other modules. Various indicators and analysis examples are as described above.

The analysis device may store kernel module information, CG gene information, and CGX inheritance levels generated by analyzing normal and tumor tissues in a separate reference DB.

59 is an example of a process 800 of analyzing a sample based on the kernel module of the genome module network. First, the reference DB may be constructed similarly to the process described in FIG. 58. The analysis device acquires a first gene expression data set of tumor tissue from the tumor tissue data DB, and acquires a second gene expression data set of enteric tissue from the normal tissue data DB. The analysis device constructs a genome module network using the first gene expression data set and the second gene expression data set, and analyzes the kernel module to obtain information on gene information, CG, CGX, etc. belonging to the kernel module as described above. Can be determined (810). The reference DB can store the result of analyzing the kernel module. The reference DB can store gene expression data sets for each normal tissue and tumor tissue, genomic module network configuration information for each normal tissue and tumor tissue, kernel configuration information for each normal tissue and tumor tissue, CG gene information, and CGX gene information. have.

The sample data DB holds gene expression data to be analyzed. An analysis target may be a patient, a normal or a candidate that falls into the normal category but is likely to develop a tumor.

The analysis device identifies kernel genes, CG genes, and CGX genes belonging to the kernel module based on the information stored in the reference DB (820).

The analysis device can perform various analysis on the sample. Hereinafter, it is assumed that the analysis device identifies gene expression data of normal tissue, gene expression data of tumor tissue, and gene expression data of a sample.

(1) The analysis device can construct a genome module network based on the gene expression data of the sample. The analysis device may determine the kernel module in the sample's genomic module network. The analysis device may determine whether the sample is a normal category or a tumor based on whether CG is present in the kernel module of the sample.

(2) The analysis device can combine gene expression data of a sample with a gene expression data set of a normal tissue, and build a genome module network based on the combined data. The analysis device can analyze the sample by comparing the sample with the normal tissue. For example, the analysis device may compare the relative entropy of CG and CGX (830), compare the MSPs of CG and CGX (840), compare the connectivity between the kernel module and other modules (850), and clustering based on MSPs. The normal and the sample may be compared by performing (860), or the normal and the sample may be compared by calculating the LOR of the clustered group (860).

(3) The analysis device can combine gene expression data of a sample with a gene expression data set of a tumor tissue, and build a genome module network based on the combined data. The analysis device can analyze the sample by comparing the tumor tissue with the sample. For example, the analysis device may compare the relative entropy of CG and CGX (830), compare the MSPs of CG and CGX (840), compare the connectivity between the kernel module and other modules (850), and clustering based on MSPs. The tumor and the sample may be compared by performing (860), or the tumor and the sample may be compared by calculating the LOR of the clustered group (860).

(4) The analysis device may combine gene expression data of a sample with gene expression data of normal tissue and gene expression data of tumor tissue, and construct a genome module network based on the combined data. The analysis device can analyze the sample by comparing the tumor tissue with the sample. For example, the analysis device may compare the relative entropy of CG and CGX (830), compare the MSPs of CG and CGX (840), compare the connectivity between the kernel module and other modules (850), and clustering based on MSPs. The normal/tumor and the sample may be compared by performing (860), or the normal/tumor and the sample may be compared by calculating the LOR of the clustered group (860).

60 is an example of a system 900 that analyzes a sample based on a kernel module of a genome module network. 60 corresponds to a system that provides a service for analyzing samples according to a sample analysis method based on a kernel module. The sample analysis method may be at least one of various methods described in FIG. 59 and the like.

The analysis system 900 includes a reference DB 910 and an analysis device 920 and/or 930. The analysis device 920 corresponds to an analysis server on a network, and the analysis device 930 corresponds to a computer device used by an individual. The computer device may be implemented in various forms such as a PC or a smart device.

The gene data generating device 80 corresponds to a device for generating gene expression data by analyzing a sample of a sample. Gene expression data can be obtained through microarray. Furthermore, gene expression data may also be prepared through NGS analysis.

The analysis device 920 may receive gene expression data of a sample through a network. In addition, the analysis device 920 includes gene expression data sets for each of the normal tissues and tumor tissues from the reference DB 910 through the network, information on the genomic module network configuration information for each of the normal tissues and tumor tissues, and the kernel configuration for each of the normal tissues and tumor tissues. Information, CG gene information, and CGX gene information may be received. The analysis device 920 may construct a genome module network using reference data and gene expression data of the sample, and analyze the sample based on the kernel module as described above. Meanwhile, the analysis device 920 may transmit the analysis result to the user terminal 50.

The analysis device 930 may acquire gene expression data of a sample through a network or a storage medium (USB, SD card, hard disk, etc.). In addition, the analysis device 930 includes gene expression data sets for each of the normal tissues and tumor tissues from the reference DB 910 through the network, information on the genomic module network configuration information for each of the normal tissues and tumor tissues, and the kernel configuration for each of the normal tissues and tumor tissues. Information, CG gene information, and CGX gene information may be received. Alternatively, unlike FIG. 60, the analysis device 930 may be connected to a reference DB by wire, or the analysis device 930 may include a reference DB in a storage medium. The analysis device 930 may build a genome module network using reference data and gene expression data of the sample, and analyze the sample based on the kernel module as described above. Meanwhile, the analysis device 930 may output the analysis result on the screen. Alternatively, the analysis device 930 may transmit the analysis result to the user terminal 50.

61 is an example of an analysis apparatus 1000 that analyzes a sample based on a kernel module of a genome module network. 61 may be a configuration of the aforementioned analysis devices 920 and 930.

The analysis device 1000 may generate health information for a sample by using a program that analyzes sample data. The analysis device 1000 may be physically implemented in various forms. For example, the analysis device 1000 may have a form such as a PC, a smart device, a computer device, a server of a network, and a chipset dedicated to data processing.

The analysis device 1000 may include a storage device 1010, a memory 1020, an operation device 1030, an interface device 1040, a communication device 1050, and an output device 1060.

The storage device 1010 may store a program constituting a genome module network and/or a program for analyzing a sample based on a kernel module of the genome module network.

The storage device 1010 may store the received gene expression data.

The storage device 1010 may store reference data obtained by analyzing normal tissues and tumor tissues. The reference data include gene expression data sets for each of normal and tumor tissues, genomic module network configuration information for each of normal tissues and tumor tissues, kernel configuration information for each of normal and tumor tissues, CG gene information, and CGX gene information. I can.

The memory 1020 may store data required for a data processing process by the analysis apparatus 1000 and temporary data generated.

The interface device 1040 is a device that receives certain commands and data from the outside. The interface device 1040 may receive gene expression data and/or reference data of a sample from an input device physically connected or an external storage device. The interface device 1040 may receive a program for processing data.

The communication device 1050 refers to a component that receives and transmits certain information through a wired or wireless network. The communication device 1050 may receive gene expression data and/or reference data of a sample from an external object. The communication device 1050 may receive a program and data for data processing. The communication device 1050 may receive reference data by communicating with a reference DB existing on a network. Meanwhile, the communication device 1050 may transmit an analysis result for the sample to the outside.

The communication device 1050 to the interface device 1040 are devices that receive certain data or commands from the outside. The communication device 1050 to the interface device 1040 may be referred to as an input device.

The output device 1060 is a device that outputs certain information. The output device 1060 may output an interface required for a data processing process and an analysis result.

The computing device 1030 can construct a genome module network using gene expression data. The computing device 1030 may build a genome module network for each of a normal tissue, a tumor tissue, and a sample.

The computing device 1030 may analyze a sample by constructing a genome module network using gene expression data of a sample and comparing kernel module information of a normal tissue (or tumor tissue) with kernel module information of the sample.

The computing device 1030 constructs a genome module network using gene expression data of a normal tissue (and/or a tumor tissue) and gene expression data of a sample, and uses the analysis method (indicator, clustering, etc.) described in FIG. The sample can be analyzed. For example, the computing device 1030 may derive a normal state, a tumor onset state, a tumor onset possibility, and a customized treatment method for a tumor as an analysis result of the sample.

The computing device 1030 may be a device such as a processor, an AP, or a chip in which a program is embedded that processes data and processes certain operations.

In addition, the genome module network, the inter-module network, the kernel module or the sample analysis method based on genes of the kernel module as described above may be implemented as a program (or application) including an executable algorithm that can be executed on a computer. The program may be provided by being stored in a temporary or non-transitory computer readable medium.

The non-transitory readable medium refers to a medium that stores data semi-permanently and can be read by a device, not a medium that stores data for a short moment, such as a register, cache, or memory. Specifically, the above-described various applications or programs include CD, DVD, hard disk, Blu-ray disk, USB, memory card, read-only memory (ROM), programmable read only memory (PROM), and eraseable PROM (EPROM). Alternatively, it may be provided by being stored in a non-transitory readable medium such as an EEPROM (Electrically EPROM) or a flash memory.

Temporary readable media are static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), and enhanced SDRAM (Enhanced SDRAM). SDRAM, ESDRAM), Synchronous DRAM (Synclink DRAM, SLDRAM) and Direct Rambus RAM (Direct Rambus RAM, DRRAM) refers to a variety of RAM.

The present embodiment and the accompanying drawings are merely illustrative of some of the technical ideas included in the above-described technology, and those skilled in the art will be able to easily within the scope of the technical ideas included in the specification and drawings of the above-described technology. It will be apparent that all of the modified examples and specific embodiments that can be inferred are included in the scope of the rights of the above-described technology.

Claims (14)

Constructing a genome module network for the sample based on entropy by the analysis device using the gene expression data of the sample; And
The analysis device comprises the step of performing an analysis on the sample based on the reference kernel module of the reference genome module network and the sample kernel module of the genome module network of the sample,
The reference genome module network is constructed in advance using at least one of a gene expression data set of a normal tissue and a gene expression data set of a tumor tissue, and the kernel module has an entropy of a reference value compared to other modules in the genome module network. A method of analyzing a sample based on a kernel module of a genome module network, which is a module having an abnormally low level, and wherein the entropy indicates a correlation between the plurality of genes based on the possibility of transcriptional expression for the plurality of genes. The method of claim 1,
The analysis device constructs the genome module network including a plurality of genome modules using gene expression data of the sample,
The step of distinguishing the plurality of genomic modules
Dividing the plurality of genes into a plurality of arbitrary sets, and adjusting the entropy of each set to be less than a threshold value while removing one gene from each of the plurality of sets; And
A sample based on a kernel module of a genome module network comprising the step of adding a gene that does not belong to the set to a set provided that the entropy for each of the plurality of sets is less than the threshold value and at the same time, the fluctuation of the reference vector is less than the reference value Analysis method. The method of claim 1,
The analysis device includes a first gene group consisting of genes present in the kernel module of the normal tissue and not present in the kernel module of the tumor tissue, and genes excluding genes belonging to the first gene group in the kernel module. A sample analysis method based on a kernel module of a genome module network comparing the reference kernel module and the sample kernel module based on at least one gene group of the second gene group. The method of claim 3,
The analysis device includes the reference kernel module and the sample based on a relative entropy between the first gene group and the second gene group, and a degree of mutation for at least one of the first gene group and the second gene group. A sample analysis method based on kernel modules of a genome module network comparing kernel modules. The method of claim 3,
The analysis device includes the reference kernel module and the sample based on a relative entropy between the first gene group and the second gene group, and a degree of mutation for at least one of the first gene group and the second gene group. A sample analysis method based on a kernel module of a genomic module network for classifying a kernel module and calculating a log odds ratio (LOR) for the reference kernel module and the sample kernel module in the classified group. The method of claim 1,
The analysis device is a sample analysis method based on a kernel module of a genome module network for comparing the connectivity between a kernel module and other modules in the reference genome module network and a kernel module and other modules in the genome module network of the sample. Constructing a genome module network based on entropy using gene expression data obtained by combining the reference gene expression data and the gene expression data of the sample by the analysis device; And
The analysis device comprises the step of performing an analysis on the sample based on the kernel module of the genome module network,
The reference gene expression data includes at least one set of a gene expression data set of a normal tissue and a gene expression data set of a tumor tissue, and the kernel module is a module having an entropy lower than a reference value compared to other modules in the genome module network. , The entropy is a sample analysis method based on a kernel module of a genomic module network indicating a correlation between the plurality of genes based on the possibility of transcriptional expression for a plurality of genes. The method of claim 7,
The analysis device constructs the genome module network including a plurality of genome modules using the combined gene expression data,
The step of distinguishing the plurality of genomic modules
Dividing the plurality of genes into a plurality of arbitrary sets, and adjusting the entropy of each set to be less than a threshold value while removing one gene from each of the plurality of sets; And
A sample based on a kernel module of a genome module network comprising the step of adding a gene that does not belong to the set to a set provided that the entropy for each of the plurality of sets is less than the threshold value and at the same time, the fluctuation of the reference vector is less than the reference value Analysis method. The method of claim 7,
The analysis device includes a first gene group consisting of genes present in the kernel module of the normal tissue and not present in the kernel module of the tumor tissue, and genes excluding genes belonging to the first gene group in the kernel module. A sample analysis method based on a kernel module of a genomic module network comparing a kernel module of the reference gene expression data with a kernel module of the sample based on at least one gene group of the second gene group. The method of claim 9,
The analysis device includes a kernel of the reference gene expression data based on a relative entropy between the first gene group and the second gene group, and a degree of mutation for at least one of the first gene group and the second gene group. A sample analysis method based on a kernel module of a genome module network comparing a module with a kernel module of the sample. The method of claim 9,
The analysis device includes the reference kernel module and the sample based on a relative entropy between the first gene group and the second gene group, and a degree of mutation for at least one of the first gene group and the second gene group. A sample analysis method based on a kernel module of a genome module network for classifying a kernel module and calculating a log odds ratio (LOR) for the reference gene expression data and the gene expression data of the sample in the classified group. The method of claim 7,
The analysis device is a sample analysis method based on a kernel module of a genome module network comparing the reference gene expression data with the gene expression data of the sample based on the connectivity between a kernel module and other modules in the genome module network. An input device for receiving reference data and gene expression data of a sample;
A storage device for storing a program for analyzing data based on a kernel module of a genome module network constructed from a gene expression data set; And
Constructing a genome module network using the gene expression data of the sample using the program, and comprising a computing device for analyzing the sample based on gene information constituting the kernel module of the constructed genome module network,
The data of the reference is at least one of a gene expression data set of a normal tissue and a gene expression data set of a tumor tissue, or a reference genome module network data constructed using the at least one set, and the kernel module is the genome module A module in which the entropy is lower than a reference value compared to other modules in the network, and the entropy is an analysis device that indicates a correlation between the plurality of genes based on the possibility of transcriptional expression for the plurality of genes. The method of claim 13,
The computing device is composed of a first gene group composed of genes present in the kernel module of the normal tissue and not present in the kernel module of the tumor tissue, and genes excluding genes belonging to the first gene group in the kernel module. Analysis device for comparing the reference and the sample based on at least one gene group among the second gene group.

Images