KR102236194B1 - Method for selecting function group marker of genes, system and method for disease prediction - Google Patents

Abstract

A method of operating a disease discrimination system, comprising: collecting functional groups in which genes having similar functions are grouped, extracting multi-functional genes with a frequency greater than or equal to a standard appearing in the functional groups, and the multi-functional genes based on functional similarity between multi-functional genes Searching for at least one common functional group index consisting of a combination of them, creating a base ontology having each common functional group index as a node, expanding the scope of the group index search with all the collected genes, and determining the functional similarity between genes. A step of creating a functional ontology by searching for at least one detailed functional group index composed of a combination of genes on the basis, adding each detailed functional group index as a node on the basis of the base ontology, and functioning the nodes constituting the functional ontology. Including the step of selecting as group indicators.

Description

Methods of selecting functional group indicators of genes, disease identification system, and operation method thereof {METHOD FOR SELECTING FUNCTION GROUP MARKER OF GENES, SYSTEM AND METHOD FOR DISEASE PREDICTION}

The present invention relates to bioinformatics technology.

Various molecular indicators such as mRNA and protein are used to find out whether a patient has a disease, a state within a disease, or a mechanism that causes disease. In recent years, in order to find an indicator that can more accurately and consistently determine a disease state, various ohmic data for each disease state are used to discover molecular indicators showing specific patterns. Ohmics data is a measure of the amount of variation or expression of all genes in a cell.By comparing all genes that may appear in a cell, a fair selection that is not biased to a specific gene and the association with the disease mechanism are comprehensively interpreted. Offers possible advantages.

Representative ohmic data include genomes, transcripts, proteins, and metabolites. Genomic mutation data comes from millions of mutation information across human DNA, and there is a very big limit in finding accurate associations in a small number of patient data. While proteomic or metabolite omics data has a great advantage in direct linkage with cell function, the number of measurable genes is still several hundred to 1,000, and it is a major disadvantage that it is limited compared to transcriptomes spanning tens of thousands of genes. . The transcriptome can be measured for the entire gene, and at the same time, it is used to measure the expression of the gene in the mRNA state. The mechanism by which the disease state is determined is complexly composed of unit functions within a cell and higher functions in which they are combined. Intracellular functions include the process by which genes informationd in DNA are expressed as proteins via mRNA, and processes in which various metabolites and cell functions in the next step are determined by the unit functions of the proteins. It is understood that the organized and networked regulatory relationships of genes that determine unit functions and combined higher functions are constituted by primary expression products (mRNA, protein) and secondary products of genes. However, at present, the causal relationship between individual biomolecules and these functions or between biomolecules and disease has been very limited. Therefore, various informational methods and modeling methods are applied to make the most efficient and accurate prediction based on available individual genes or comprehensive ohmic experimental data. Transcript analysis data, such as DNA microarray or RNASeq, has the largest range of identifiable biomolecules for all genes, and provides data on a wide variety of patients with convenience of measurement. Therefore, it can be said that the transcriptome analysis data is the most efficient resource that can predict the causal relationship between the disease mechanism and the biomolecule by reflecting the complexity of the biomechanism.

In the selection of biomolecular indicators for achieving clinical disease diagnosis, prognosis prediction, and drug companion diagnosis, the accuracy and stability/reproducibility of discrimination can be secured by applying a technique that reflects organized and networked biological mechanisms. Currently, most disease transcripts are used to select genes whose expression levels of mRNA (gene transcription substances) significantly change according to disease states, and from them, index them using an expression index of an individual gene or a combination of genes. . Since the disease state is a result of the synthesis of various gene functions, indexing using the difference in expression of individual genes has a fundamental limitation, and patients with the same disease state also show large deviations, so the reliability and stability/reproducibility of the selected index are greatly degraded. It is difficult to distinguish the difference between mutations that are related to the disease state of individual patients and mutations that are not related to it.

By integrating and expressing the mutations of the genes in the functional group that make up the disease mechanism, mutations that are not related to the disease state can be offset and the related mutations can be reinforced, resulting in stable indicators covering various patients with the same disease state. . As a technique that has been recently tried, several individual molecular indicators with similar expression patterns are grouped into a single group, or genes within known functional groups are grouped and the final group is formed according to their expression pattern similarity. .

However, since the group indexing based on expression pattern similarity is still affected by the patient group (sample) participating in the pattern discovery, it is improved compared to the individual genetic index, but the accuracy and stability are not significantly improved. It is difficult to determine functional groups similar to these pattern groups, and it is difficult to determine whether these pattern groups are meaningful until the number of patient samples increases significantly.

Although the previously known functional group-centered group indexing has various functional information, there is a fundamental limitation in determining the functional group using only limited functional information represented by Gene Ontology. Currently, it is possible to derive a combination of millions of functional groups through various experimental data and information analysis data, but there is no technology that can synthesize millions of functional groups and reconstruct them into comprehensive and essential functional groups. Therefore, it is a situation in which only a single resource is used or all functional information in the resource is simply merged and analyzed.

In addition, the group index selected based on the expression pattern similarity and the group index selected based on the known functional group do not show the discrimination performance of the group index candidates in various disease states or disease mechanisms. Group indicators that have been tried to date cannot be considered as stable indicators that determine disease status based on disease mechanisms.

Therefore, a new group that overcomes the limitations of the existing individual gene expression indicators, which had poor reliability and stability due to difficulty in searching for meaningful mutations in diseases, and also overcomes the limitations of the existing functional group analysis method that did not utilize various functional information. Indicators are required.

(Patent Document 1) KR10-1927910 B

(Patent Document 2) KR10-1990429 B

(Patent Document 3) KR10-1860061 B

The task to be solved is to (re)constitute genes with similar functions into one functional group indicator, and to provide a system and method for discriminating diseases using group indicators.

The task to be solved is to provide a system and method that integrates and analyzes various functional information, reorganizes genes of comprehensive and essential functions related to disease, discovers disease group indicator candidates, and selects disease group indicators through disease discriminant verification. will be.

The task to be solved is to discover new functional group indicators by reorganizing various functional groups, to find disease group indicators by scoring disease associations of each functional group indicator, and to develop a discrimination model that can discriminate disease status based on disease group indicators. It is to provide a system and method to configure.

A method of operating a disease discrimination system according to an embodiment, comprising: collecting functional groups in which genes having similar functions are grouped, extracting multi-functional genes having a frequency of occurrence in the functional groups greater than or equal to a reference, and functional similarity between multi-functional genes. Based on the search for at least one common functional group index composed of a combination of the multifunctional genes, generating a base ontology having each common functional group index as a node, expanding the group index search range with all the collected genes, Searching for at least one detailed functional group index composed of a combination of genes based on the functional similarity between genes, and creating a functional ontology by adding each detailed functional group indicator as a node based on the base ontology, and the functional ontology And selecting the constituent nodes as functional group indicators.

In the step of generating the base ontology, the first gene network is expanded by connecting the corresponding multi-functional gene pairs in the order of multi-functional gene pairs having high functional similarity between multi-functional genes, and a maximum clique is searched in the first gene network. And, if the gene set corresponding to the searched maximum click exists in the collected functional groups, the procedure of generating the searched maximum click as a node of the ontology can be repeated. The gene set corresponding to the maximum click may be the common functional group index.

In the step of generating the base ontology, after adding each multifunctional gene pair to the first gene network, the added multifunctional gene pair expands the maximum clicks searched with the previously added multifunctional gene pairs or constructs a new maximum click. You can explore whether it is.

In the step of generating the base ontology, the gene set corresponding to the searched maximum click is determined as a node candidate, and if there is a subset node of the node candidate in the ontology, the node candidate is added as a parent node of the subset node. And, if there is no subset node of the node candidate in the ontology, the node candidate may be added as an end node.

In the step of generating the functional ontology, the function similarity between genes is calculated for the collected genes, the gene pairs having high function similarity are connected in the order of the gene pairs to expand the second gene network, and the second gene network If a maximum clique is searched for and a gene set corresponding to the searched maximum click exists in the collected functional groups, a procedure of generating the searched maximum click as a node of the ontology may be repeated. The gene set corresponding to the maximum click may be the detailed functional group index.

In the step of generating the functional ontology, after each gene pair is added to the second gene network, it is searched whether the added gene pair expands the maximum clicks searched with previously added gene pairs or constitutes a new maximum click. I can.

In the step of generating the functional ontology, a gene set corresponding to the searched maximum click is determined as a node candidate, and if there is a subset node of the node candidate in the ontology, the node candidate is added as a parent node of the subset node. And, if there is no subset node of the node candidate in the ontology, the node candidate may be added as an end node.

The operation method may further include selecting, as a disease group index, a functional group index that significantly includes a disease index among the functional group indexes, and has a significantly different activation score in the disease-non-disease microarray. .

The operation method includes generating learning data by calculating activation scores of disease group indicators for each microarray sample of a disease state to be determined, and learning a discrimination model for determining a specific disease state based on the activation scores of the disease group indicators. When a request is made to determine the disease state of a specific sample, inputting the activation score of the disease group indicators for the specific sample into the learned discrimination model, and the specific sample through the discrimination value output from the discrimination model. It may further include the step of outputting the disease state of.

In another embodiment, as an operating method of another disease determination system, collecting functional groups in which genes having similar functions are grouped, expanding a gene network by connecting gene pairs based on the similarity of the functional groups in which the collected genes are included, Searching for a functional group indicator consisting of a combination of genes connected in the gene network, generating an ontology having each functional group indicator as a node, selecting nodes constituting the ontology as functional group indicators, and the function Among the group indicators, a disease indicator is significantly included, and a functional group indicator having a significantly different activation score in the disease-non-disease microarray is selected as the disease group indicator.

The operation method includes generating learning data by calculating activation scores of disease group indicators for each microarray sample of a disease state to be determined, and learning a discrimination model for determining a specific disease state based on the activation scores of the disease group indicators. When a request is made to determine the disease state of a specific sample, inputting the activation score of the disease group indicators for the specific sample into the learned discrimination model, and the specific sample through the discrimination value output from the discrimination model. It may further include the step of outputting the disease state of.

The step of generating the ontology includes extracting genes with a frequency greater than or equal to a standard appearing in the functional groups as multi-functional genes, at least one common functional group index composed of a combination of the multi-functional genes based on the similarity of functions between multi-functional genes. And generating a base ontology having each common functional group index as a node, and expanding the search range of the group index with all the collected genes, and at least one consisting of combinations of genes based on the functional similarity between genes. And generating the ontology by searching for a detailed functional group index and adding each detailed functional group index as a node based on the base ontology.

In the step of generating the ontology, the gene network is expanded by connecting the corresponding gene pairs in the order of gene pairs having high functional similarity between genes, searching for a maximum clique in the gene network, and searching for a maximum click. The procedure of generating the corresponding gene set as a node of the ontology can be repeated.

The step of collecting the functional groups includes a database providing functional gene set information, and a database providing biological pathways included in the disease pathway, various regulator-target information, and gene interaction information. It is possible to collect functional groups in which similar genes are grouped.

As a disease discrimination system according to an embodiment, a gene network is expanded by collecting functional groups in which genes having similar functions are grouped, and by sequentially connecting gene pairs based on the similarity of the functional groups including the collected genes, and the gene A functional group indicator discovery device that selects the gene set of the maximum clique searched in the network as a functional group indicator, and a disease indicator significantly included among a plurality of functional group indicators, in a disease-non-disease microarray. A discrimination model that selects functional group indicators with significantly different activation scores as disease group indicators, calculates activation scores of disease group indicators for each microarray sample, and determines a specific disease state based on the activation scores of the disease group indicators It includes a disease discrimination model generation device for learning.

The functional group index discovery device extracts multi-functional genes with a frequency of occurrence of the functional groups greater than or equal to a reference, and expands a first gene network by connecting multi-functional gene pairs arranged based on functional similarity between multi-functional genes, and the first gene network. 1 It is possible to create a base ontology with the gene set of the maximum click searched in the gene network as a node. The functional group index discovery device arranges gene pairs based on the functional similarity between genes of all the collected genes, expands the second gene network by connecting the gene pairs in order, and maximizes clicks searched in the second gene network. The final ontology may be generated by adding the gene set of to the base ontology. The functional group indicator discovery apparatus may select nodes constituting the final ontology as functional group indicators.

When a request for determining a disease state of a specific sample is requested, the disease determination model generating device inputs activation scores of the disease group indicators for the specific sample into the learned determination model, and the determination value output from the determination model You can print the disease state of a specific sample.

According to the embodiment, genes with similar functions can be reconstructed into a comprehensive and essential functional group, and through this, it is possible to reflect a disease mechanism complicatedly composed of unit functions within a cell and higher functions in which they are combined, thereby discriminating diseases. Accuracy and stability/reproducibility can be improved.

According to an embodiment, by integrating and reconstructing a gene set previously known as a functional group, a gene set with a known causal relationship due to a function can be discovered as one "function group indicator", and in particular, a "common function group" related to various functions. By a combination of “indicators” and “sub-functional group indicators” associated with specific sub-functions, it is possible to select functional group indicators that can describe disease and cellular functions.

According to the embodiment, it is possible to select a "disease group indicator" showing a specific expression pattern in a disease while having a functional causal relationship through analysis of disease indicators and disease transcript data, and a discrimination model is constructed based on the disease group indicators, The reproducibility of disease state determination can be improved.

The disease group indicator selected according to the embodiment may be widely used for disease diagnosis, prognosis prediction, drug accompanying diagnosis, etc., and may be manufactured and used as a microarray or a multiplex analysis kit.

According to the embodiment, it overcomes the limitations of the existing individual gene expression index, which was inferior in reliability and stability due to difficulty in searching for meaningful mutations in diseases, and solves the problems of the existing functional group analysis method, which was not able to synthesize and utilize various functional information. I can.

1 is a block diagram of a disease determination system using a group index of genes similar to function according to an embodiment.
2 is a flowchart of a method for configuring a base ontology according to an embodiment.
3 is a flowchart of a method of configuring a functional ontology according to an embodiment.
4 is a flowchart of a method for selecting a disease group index according to an exemplary embodiment.
5 is a flowchart of a method for generating a disease discrimination model according to an exemplary embodiment.
6 is a diagram illustrating a method of configuring a base ontology according to an exemplary embodiment.
7 is a diagram illustrating a method of configuring a functional ontology according to an exemplary embodiment.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those of ordinary skill in the art may easily implement the present invention. However, the present invention may be implemented in various different forms and is not limited to the embodiments described herein. In the drawings, parts irrelevant to the description are omitted in order to clearly describe the present invention, and similar reference numerals are attached to similar parts throughout the specification.

Throughout the specification, when a part "includes" a certain component, it means that other components may be further included rather than excluding other components unless specifically stated to the contrary. In addition, terms such as "... unit", "... group", and "module" described in the specification mean a unit that processes at least one function or operation, which can be implemented by hardware or software or a combination of hardware and software. have.

Throughout the specification, a group marker composed of a group of genes (gene sets) having the same/similar functions is simply referred to as a “functional group indicator”. Depending on the functional range associated with the functional group indicator, it can be called as a common functional group indicator or a detailed functional group indicator. It is determined that the function is similar based on the similarity determination criterion, or information known to have similar function may be used.

1 is a block diagram of a disease determination system using a group index of genes similar to function according to an embodiment.

Referring to FIG. 1, the disease determination system 10 operates as at least one processor, a function group indicator discovery device 100 that selects genes with similar functions as function group indicators, and the disease association of the functional group indicators is evaluated. Thus, it includes a disease discrimination model generation device 200 for selecting a disease group index and generating a discrimination model based on the disease group index. The disease determination system 10 may further include a database 300 that stores information necessary for the operation of the function group indicator discovery device 100 and the disease determination model generation device 200. The group indicator discovery device 100, the disease determination model generation device 200, and the database 300 may interwork with each other through a communication interface, and may exchange information necessary for the operation of the present invention. At least a part of the database 300 may be built in the disease determination system 10 or may be implemented in an external server.

First, the database 300 can be classified into a functional gene set database 310, a pathway and interaction database 320, a disease index database 330, and a microarray database 340 according to data types. have. Naturally, the database 300 may be implemented in various ways according to design, and does not necessarily need to be physically classified, and does not necessarily need to exist in the same physical location.

The functional gene set database 310 provides relationship information between a function and a gene. The functional gene set database 310 may provide functional gene set information such as Gene Ontology, MSigDB, and Enrichr.

The pathway and interaction database 320 provides biological pathways, various regulator-target information, and gene interaction information included in the disease pathway. The pathway and interaction database 320 includes, for example, BioCarta, HumanCyc, KEGG Pathway, NCI-PID, Panther Pathway, PharmGKB, Reactome, biological pathways included in disease pathways in the SMPDB database, TF modulator-targets in the TRANSFAC database. Information, E3 modulator-target information in the E3Net database, kinase modulator-target information in the PhosphoSitePlus database, dephosphorylase modulator-target information in the DEPOD database, and gene interaction information in the HIPPIE database can be provided.

The disease indicator database 330 provides known disease indicators. The disease index database 330 may provide, for example, known disease gene information in the DisGeNet database and drug target information for a disease in the DrugCentral database.

The microarray database 340 provides microarray information. The microarray database 340 may provide, for example, gene expression microarray information in the Gene Expression Omnibus or ArrayExpress database.

The functional group indicator discovery apparatus 100 selects genes having similar functions as functional group indicators using information from the functional gene set database 310 and the pathway and interaction database 320. To this end, the function group index discovery apparatus 100 may include a gene function integrated extraction unit 110, a base ontology constructing unit 130, and a function group index selection unit 150.

The gene function integration extraction unit 110 acquires gene function integration information in which the function information of the gene included in the known function gene set and the function information of the gene included in the path and interaction information are integrated. Meanwhile, the gene function integrated extraction unit 110 may add function information obtained from a list of random genes defined by a user to the integrated gene function information. The gene function integration extraction unit 110 acquires multifunctional gene and inter-gene function similarity information from the gene function integration information.

The gene function integration extraction unit 110 may collect genes having the same/similar functions from functional-gene information such as biological functions, molecular functions, and intracellular locations into a functional group (module). In this case, the gene function integration extraction unit 110 may extract a function module from a function gene set database 310 such as Gene Ontology, MSigDB, or Enrichr.

In addition, the gene function integration extraction unit 110 may collect genes having the same path from the path information of the path and interaction database 320 into a functional group. The gene function integration extraction unit 110 does not have a pathway in the pathway and interaction database 320, but when there is interaction information about a regulator such as TF, E3, miRNA, kinase, phosphatase, etc., it is regulated by the same regulator. Genes can be grouped together and collected into functional groups. When there is only interaction information in the path and interaction database 320, the gene function integration extraction unit 110 may collect the 1-hop network of each gene into one gene set and collect it as a function group.

In addition, the gene function integration extraction unit 110 may collect a list of arbitrary gene sets defined by the user, such as a related gene set derived from ohmic data such as a microarray, and a specific expression gene set for each condition, as a function group.

The gene function integration extraction unit 110 calculates the frequency of appearance in the function groups for each gene from the gene function integration information in which the collected various function groups are integrated, and extracts a gene having a high frequency on a predetermined basis as a multifunctional gene. A certain criterion may be set to, for example, the top 5%, a p-value of 0.05 or less.

In addition, the gene function integration extraction unit 110 performs a similarity analysis by scoring whether the function group including the corresponding gene is similar to all the gene pairs included in the gene function integration information. The function similarity information between genes calculated through similarity analysis may be expressed as a function similarity matrix between genes. The gene function integration extraction unit 110 may analyze functional similarity between genes by using, for example, an analysis method of a Jacquard index.

The base ontology configuration unit 130 and the function group indicator selection unit 150 use the multifunctional gene information and the function similarity information between genes extracted from the gene function integration extraction unit 110 to select a gene set with a known causal relationship due to a function. Discovered as one "functional group indicator". Functional group indicators consist of a set of genes involved in function.

Specifically, each of the base ontology constructing unit 130 and the functional group indicator selecting unit 150 is a graph theory-based gene network that connects the relationship of gene pairs with a gene node (vertex) and a connecting line (side), and extracted from the gene network. A functional group index is selected based on a hierarchical ontology that generates a maximum clique as an ontology node. Ontology is expressed as a set of directed acyclic graphs, based on a base node composed of a small number of multifunctional gene sets, by connecting the parent (parent) node composed of a detailed function gene set related to a subdivided function, or adding a new terminal node. Is created. In ontology, an upper (parent) node contains a gene set of a lower node, and a lower node is a subset of the gene set included in an upper node.

Although the base ontology configuration unit 130 and the function group indicator selection unit 150 do not necessarily need to be independently implemented, in order to explain step by step a method of configuring a base ontology and configuring a final ontology based on the base ontology, It will be described as a separate configuration. In addition, functional group indicators can be classified into “common functional group indicators” associated with multi-functionality and “detailed functional group indicators” associated with subdivided functions, but may be referred to as functional group indicators without distinction.

The base ontology construction unit 130 constructs a base ontology based on the function similarity information of the multifunctional genes extracted by the gene function integration extraction unit 110. The base ontology constructing unit 130 extracts functional similarity information between multifunctional genes from among the functional similarity information between genes. The base ontology construction unit 130 gradually expands the gene network by adding a connection relationship while connecting the two genes with a node and a connection line in the order of gene pairs having high functional similarity. At this time, the base ontology configuration unit 130 gradually expands the network by adding a gene pair, searching for a maximum clique in which a gene node can no longer be added in the gene network, and multifunctional included in the maximum click. Genes are ontologyized as common functional group indicators. That is, the base ontology construction unit 130 generates multi-functional genes included in the maximum click as base ontology nodes. Multifunctional genes constituting the underlying ontology node are selected as common functional group indicators. At this time, the base ontology construction unit 130 determines whether a gene pair is a maximum click every step of adding a gene pair from the gene network, and the multifunctional genes included in the maximum click are among various functional groups collected by the gene function integration extraction unit 110. In the case of more than one subset, multifunctional genes included in the maximum click can be generated as ontology nodes, which are common functional group indicators.

The functional group index selection unit 150 expands the base ontology constructed by using the multifunctional gene pair in the base ontology construction unit 130 by using the entire gene pair. The functional group index selection unit 150 expands the gene network to the entire gene pair and adds the genes corresponding to the maximum click to the upper node of the base ontology node in the same manner as the base ontology configuration method of the base ontology configuration unit 130 do. Genes constituting an upper ontology node are selected as detailed functional group indicators related to a detailed function than a lower ontology node or a base ontology node. Meanwhile, since the functional group index selection unit 150 expands the gene network to the entire gene pair, a functional group of genes other than the multi-functional gene may be generated. In this case, the functional group indicator selection unit 150 adds a functional group of genes other than the multifunctional gene to the ontology as a detailed functional group indicator.

In this way, the functional group indicator selection unit 150 calculates functional similarity information between genes obtained by integrating the functional information of various genes and scoring whether the functional group including the gene is similar, and using this to ontology the functional group indicator. . Accordingly, the functional group indicator selection unit 150 may extract various functional group indicators from common functional group indicators to detailed functional group indicators by reconstructing various functional gene sets.

The disease discrimination model generation apparatus 200 includes a disease group indicator selection unit 210 that selects a disease group indicator by evaluating the disease association of the functional group indicators, and a disease state discriminator that generates a discrimination model based on the disease group indicator ( 230).

The disease group index selection unit 210 significantly includes the disease index among the functional group indexes, evaluates that the difference in expression in the microarray data is significant, and selects the disease group index based on the evaluation result. In more detail, the disease group index selection unit 210 may filter a functional group index that significantly includes a disease index (disease marker) extracted from the disease index database 330 among all functional group indexes as a disease group index candidate. In addition, the disease group index selection unit 210 calculates the activation score of the disease group index candidates for each sample based on the microarray gene expression data extracted from the microarray database 340, and shows a significant difference between the disease states to be checked. A candidate disease group indicator can be selected as the final disease group indicator. The disease group indicator consists of a disease-related gene set.

The disease state discriminator 230 generates learning data by calculating activation scores of disease group indicators for each microarray sample of the disease state to be determined, and can determine a specific disease state based on the activation scores of the disease group indicators. Train the discriminant model. The discriminant model may be a support vector machine (SVM) machine learning discriminator capable of discriminating a disease state in microarray data, but various learning models may be used.

When the disease state discriminator 230 receives a request to determine the disease state of a new sample, it calculates activation scores of the disease group indicators for the new sample, and inputs this into the learned discrimination model. Then, the disease state determiner 230 determines the disease state of the new sample through the determination value output from the determination model. The disease state discriminator 230 replaces the gene expression value in the sample with the activation score of the disease group indicators used in the learning of the discriminant model, and uses this as an input value of the discrimination model.

In this way, the disease determination system 10 extracts gene function integration information from functional information such as various functional gene sets, pathways, and interaction information, and extracts functional group indicators based on the similarity of functions between genes. At this time, the functional group indicator includes a common functional group indicator and a detailed functional group indicator, and the disease identification system 10 first constructs a basic ontology consisting of common functional group indicators using multifunctional genes, and then uses all genes to form a detailed functional group. By exploring the indicators, the underlying ontology is expanded. The disease determination system 10 selects a disease group index related to a disease based on known disease index and microarray data from among the function group indexes obtained by integrated analysis of various function information, and then uses the selected disease group index to determine the disease. It trains a discrimination model capable of discriminating the state.

As described above, the disease state discriminator of the present invention selects a disease-related gene set defined as a disease group index, and constructs a discrimination model with a combination thereof. Accordingly, the present invention can construct a more robust discrimination model by solving the limitation of reproducibility of a conventional single marker or marker set. In addition, the present invention utilizes group information that has not been utilized in the conventional gene set selection method, so that it is possible to directly link with known function information, so that mechanism analysis is possible at the same time.

2 is a flowchart of a method for configuring a base ontology according to an embodiment.

Referring to FIG. 2, the function group indicator discovery apparatus 100 uses genes (gene sets) having the same/similar functions, pathways, or interactions as a functional group based on various functional gene sets and pathways and interaction information. Bundled and extracted gene function integration information (S110).

The functional group index discovery apparatus 100 extracts multifunctional genes from the gene function integration information in which various functional groups are integrated (S120). Multifunctional genes can be determined by information on the frequency of appearance in functional groups.

The functional group index discovery device 100 calculates the functional similarity between the multifunctional genes, and arranges the multifunctional gene pairs in the order of the highest functional similarity (S130). The functional similarity between genes is calculated by a similarity analysis by scoring whether a functional group containing a corresponding gene is similar for a gene pair, and can be expressed as a functional similarity matrix between genes.

The functional group index discovery apparatus 100 adds gene pairs to the gene network in the order of high functional similarity among the multi-functional gene pairs, and searches for a maximum click in the gene network (S140). To this end, the functional group index discovery device 100 lists all multifunctional gene pairs as candidates for an additional connection relationship, then creates an empty network to be expanded by adding a connection relationship, and step by step from a gene pair having a high functional similarity. The gene network is expanded while adding. On the other hand, searching for a maximum click each time a gene pair is added to the network takes a lot of computation time. Therefore, the functional group index discovery apparatus 100 calculates only whether a new gene added to the gene network can expand existing maximum clicks or constitute a new maximum click based on the maximum clicks found in the previous step. . In order to shorten the calculation time, the functional group index discovery apparatus 100 may perform parallel calculation and search for maximum clicks.

When a new maximum click is found in the gene network, the functional group indicator discovery apparatus 100 determines a gene set corresponding to the maximum click as a common functional group indicator, and selects the common functional group indicator as an ontology node candidate (S150). The functional group index discovery device 100 exists in the gene function integration information collected by the functional group including all the gene sets corresponding to the maximum click, or the maximum click can be configured by a combination of nodes (gene sets) in the ontology. In this case, the gene set of the maximum click may be selected as an ontology node candidate.

The functional group index discovery apparatus 100 searches whether there is a node in the ontology that is a subset of the ontology node candidate (S160). That is, the functional group index discovery apparatus 100 searches for a node composed of a subset of a gene set constituting an ontology node candidate.

If there is a node that is a subset of the ontology node candidate in the ontology, the functional group index discovery apparatus 100 adds the ontology node candidate as a parent node of the subset node, and if there is no node that is a subset of the ontology node candidate, the ontology node candidate Is added as a new end node (S170).

The functional group index discovery device 100 determines whether a gene pair to be added to the gene network remains among all multi-functional gene pairs (S180). The functional group index discovery apparatus 100 may list all multifunctional gene pairs as candidates for an additional linkage, and then determine whether a gene pair to be added to the gene network remains while removing the searched gene pair by adding it to the gene network. The functional group index discovery apparatus 100 repeats the step (S140) of searching for gene network expansion and maximum click when there are remaining gene pairs to be added to the gene network.

When the linking of all the gene pairs of the multi-functional genes is completed, the functional group indicator discovery apparatus 100 outputs a base ontology in which each common functional group indicator is composed of nodes (S190). The functional group index discovery apparatus 100 constructs nodes of a base ontology by searching for a gene set functionally connected between multifunctional genes.

3 is a flowchart of a method of configuring a functional ontology according to an embodiment.

Referring to FIG. 3, the functional group indicator discovery apparatus 100 expands the basic ontology generated by multifunctional genes to the entire gene to construct a functional ontology. The functional group indicator discovery apparatus 100 searches for a detailed functional group indicator composed of a gene set related to a detailed function from the common functional group indicator of the underlying ontology. The method of constructing the functional ontology is similar to the method of constructing the base ontology.

The functional group index discovery device 100 receives a base ontology generated by multifunctional genes (S210).

The functional group index discovery apparatus 100 calculates a function similarity between all genes from the gene function integration information in which various functional groups are integrated (S220). The functional similarity between genes is calculated by a similarity analysis by scoring whether a functional group containing a corresponding gene is similar for a gene pair, and can be expressed as a functional similarity matrix between genes. Functional similarity between all genes can be calculated in advance.

The functional group index discovery apparatus 100 sorts all gene pairs in the order of the highest functional similarity (S230).

The functional group index discovery apparatus 100 adds gene pairs to the gene network in the order of high functional similarity among all gene pairs, and searches for a maximum click in the gene network (S240). To this end, the functional group index discovery device 100 lists all gene pairs as candidates for additional connection relations, then creates an empty network to be expanded by adding the connection relations, and stepwisely selects nodes from gene pairs having high functional similarity. As you add, you expand your genetic network. At this time, the functional group index discovery device 100 calculates only whether a new gene added to the gene network can expand existing maximum clicks or constitute a new maximum click based on the maximum clicks found in the previous step. You can shorten the search time. In addition, the functional group index discovery apparatus 100 may perform parallel calculations and search for maximum clicks in order to shorten the calculation time.

When a new maximum click is found in the gene network, the functional group indicator discovery apparatus 100 determines a gene set corresponding to the maximum click as a detailed functional group indicator, and selects the detailed functional group indicator as an ontology node candidate (S250). The functional group index discovery device 100 exists in the gene function integration information collected by the functional group including all the gene sets corresponding to the maximum click, or the maximum click can be configured by a combination of nodes (gene sets) in the ontology. In this case, the gene set of the maximum click may be selected as an ontology node candidate.

The functional group index discovery apparatus 100 searches for a node that is a subset of the ontology node candidate in the ontology (S260).

If there is a node that is a subset of the ontology node candidate in the ontology, the functional group index discovery apparatus 100 adds the ontology node candidate as a parent node of the subset node or adds the ontology node candidate as a new end node (S270). Through this, the concept of a higher functional group is built up based on a basic ontology composed of multifunctional genes, but as a gene set containing genes other than multifunctional genes is added as a functional group, group indicators of detailed functions can be searched.

The functional group index discovery apparatus 100 determines whether a gene pair to be added to the gene network remains among all gene pairs (S280). After listing all the gene pairs as candidates for an additional linkage relationship, the functional group index discovery apparatus 100 may determine whether a gene pair to be added to the gene network remains while removing the searched gene pair by adding it to the gene network. If there is a gene pair to be added to the gene network, the functional group index discovery device 100 repeats the step (S240) of searching for a gene network expansion and a maximum click.

When the entire gene pair is added to the gene network, the functional group index discovery apparatus 100 outputs the nodes of the functional ontology as functional group indexes (S290). The gene set constituting each node of the functional ontology constitutes a functional group indicator, and includes a common functional group indicator and a detailed functional group indicator.

4 is a flowchart of a method for selecting a disease group index according to an exemplary embodiment.

Referring to FIG. 4, the disease discrimination model generation apparatus 200 significantly includes a known disease indicator (marker) among functional group indicators, and a functional group indicator in which scores are significantly different in disease-non-disease microarray data. Is selected as the disease group indicator.

The apparatus 200 for generating a disease discrimination model selects a functional group indicator that significantly includes a known disease indicator (marker) among functional group indicators as a disease group indicator candidate (S310). The apparatus 200 for generating a disease discrimination model may extract a known disease index from the disease index database 330. To analyze the significance of disease indicators, for example, by using Fisher's exact test, the significance indicating the degree of disease indicator inclusion is calculated for each functional group indicator, and the functional group indicator with a p-value of 0.05 or less is the disease. Can be selected as a group indicator candidate.

The disease discrimination model generation apparatus 200 calculates activation scores of disease group index candidates for each sample based on the microarray data (S320). The disease discrimination model generation apparatus 200 selects a functional group index having a significantly different activation score in the disease-non-disease microarray data as a disease group index. The apparatus 200 for generating a disease discrimination model may integrate the gene expression values of the disease group index candidate, and calculate an activation score of the disease group index candidate based on this. The disease discrimination model generation apparatus 200 converts gene expression information expressed in a disease state sample into an activation score of a disease group indicator candidate of the sample through Gene Set Enrichment Analysis. That is, through gene set enrichment analysis, the sample-gene expression matrix is transformed into an activation score matrix of the sample-disease group indicator candidate. For the gene set enrichment analysis, for example, single sample GSEA, gene set variation analysis, etc. can be used. Microarray data may utilize gene expression microarray information in, for example, Gene Expression Omnibus or ArrayExpress databases, and may be extracted from the microarray database 340.

The disease discrimination model generation apparatus 200 compares the difference in activation score between disease states for each disease group index candidate, and selects a specific disease group index candidate showing a significant difference as a disease group index of the corresponding disease state (S330). The disease discrimination model generation apparatus 200 analyzes the difference in activation score between disease states through the activation score matrix of the sample-disease group index candidate, and selects a specific group index candidate showing a significant difference as the disease group index. To analyze the significance of the disease group index candidate, for example, a t-test may be used to calculate the significance of the difference in activation score, and a group index with a p-value of 0.05 or less may be selected as the disease group index.

5 is a flowchart of a method for generating a disease discrimination model according to an exemplary embodiment.

Referring to FIG. 5, the apparatus 200 for generating a disease determination model calculates an activation score of a disease group index for each microarray sample of a disease state to be determined (S410). The apparatus 200 for generating a disease discrimination model may calculate an activation score by converting the sample-gene expression matrix into an activation score matrix of the sample-disease group index.

The disease discrimination model generating apparatus 200 learns a discrimination model capable of discriminating a specific disease state based on the activation score of the disease group indicators (S420). The discriminant model may be a support vector machine (SVM) machine learning discriminator capable of discriminating a disease state in microarray data, but various learning models may be used.

The disease determination model generation apparatus 200 receives a new sample for determining a disease state (S430).

The disease discrimination model generating apparatus 200 converts the gene expression value (gene expression matrix) of the input sample into activation score information (activation score matrix) of the disease group indicators used for learning the discrimination model (S440).

The disease discrimination model generating apparatus 200 inputs activation score information of the input sample as the learned discrimination model (S450).

The disease discrimination model generation apparatus 200 outputs a discrimination value of a discrimination model for a new sample (S460). The disease state of the input sample is determined through the discrimination value.

6 is a diagram illustrating a method of configuring a base ontology according to an exemplary embodiment.

With reference to FIGS. 2 and 6, the function group index discovery device 100 adds the gene pairs with high similarity one by one from among the functional similarity information between multifunctional genes extracted by the gene function integration extraction unit 110 to the gene network, while maximizing clicks. How to find and add to the ontology will be described as an example.

For example, when the functional group index discovery device 100 finds six genes of MDM2, p53, BAX, CASP8, BCL2, and CDK2 as multi-functional genes, it calculates functional similarity between multi-functional genes as shown in Table 1. According to the functional similarity of Table 1, multifunctional gene pairs having high functional similarity can be arranged as shown in Table 2.

MDM2 p53 BAX CASP8 BCL2 CDK2 MDM2 - 0.236 0.190 0.125 0.105 0.163 P53 0.236 - 0.195 0.137 0.124 0.165 BAX 0.190 0.195 - 0.200 0.186 0.168 CASP8 0.125 0.137 0.200 - 0.187 0.100 BCL2 0.105 0.165 0.186 0.187 - 0.092 CDK2 0.163 0.124 0.168 0.100 0.092 -

One MDM2-p53 0.236 2 BAX-CASP8 0.200 3 p53-BAX 0.195 4 MDM2- BAX 0.190 5 CASP8- BCL2 0.187 6 BAX- BCL2 0.186 7 BAX- CDK2 0.168 … …

It describes the construction of the underlying ontology using “known function 1: p53 pathway” and “known function 2: apoptosis”. The p53 pathway consists of {MDM2, p53, BAX}, and apoptosis consists of {BAX, CASP8, BCL2}, and is assumed to exist in the gene function integration information.

In step 1, the functional group index discovery apparatus 100 generates an empty gene network with no relationship and an empty ontology without ontology notes.

In step 2, the functional group index discovery apparatus 100 adds the relationship of MDM2-p53 having the highest functional similarity in Table 2 to the gene network, and searches for a maximum click. At this time, {MDM2, p53}, which consists of two genes, is the newly searched maximum click. The functional group index discovery apparatus 100 checks whether the maximum click searched for in the collected gene integration function information exists, and if it exists in the gene integration function information, adds it to the ontology as a new node. Since {MDM2, p53} shares “Known Function 1: P53 pathway”, it is added to the ontology as a new ontology node (T1).

In step 3, the functional group index discovery apparatus 100 adds BAX-CASP8, which has the second highest functional similarity in Table 2, to the network. Since {BAX, CASP8} is a new extreme click and shares “Known Function 2: Apoptosis”, it is added to the ontology as a new ontology node (T2).

In step 4, {p53, BAX}, which has the third highest functional similarity, is a new maximum click, and is added as an ontology node T3.

In step 5, MDM2-BAX, which has the fourth highest functional similarity, is added to the gene network, and {MDM2, p53, BAX} is a new maximal click. All three genes share the “known function 1: P53 pathway”, while ontology node T1 (MDM2, p53) and ontology node T3 (p53, BAX) are subsets of {MDM2, p53, BAX}, so {MDM2, p53 , BAX} is added as a new ontology node T4, which is the parent node of T1 and T3.

In step 6, {CASP8, BCL2} having the fifth highest functional similarity is added as a new ontology node T5.

In step 7, the maximum click {BAX, BCL2, CASP8} found by {BAX, BCL2} with the sixth highest functional similarity is added as a new ontology node (T6), which is the parent node of the ontology node T2 and the ontology node T5. .

In step 8, BAX-CDK2 with the seventh highest functional similarity is added to the gene network, and {BAX, CDK2} is searched for with a new maximum click. However, because {BAX, CDK2} is not included in “known function 1: p53 pathway” and “known function 2: apoptosis”, we judge that it does not have shared (common) function. Therefore, {BAX, CDK2} is not added to the ontology.

As such, by adding genetic relationships one by one, ontology expansion is performed until all relationships are displayed in the gene network, and the finally obtained ontology can be obtained as a base ontology. The base ontology is created with nodes composed of multifunctional gene pairs, as shown in Table 3 below. In this process, relationships between genes are added one by one for explanation, but an ontology can be constructed by adding a relationship set by a specific number or at a specific similarity score interval.

order Multifunctional gene pair Functional similarity Ontology Node One MDM2-p53 0.236 T1 2 BAX-CASP8 0.200 T2 3 p53-BAX 0.195 T3 4 MDM2- BAX 0.190 T4={ T1, T3} 5 CASP8- BCL2 0.187 T5 6 BAX- BCL2 0.186 T6={T2, T5} 7 BAX- CDK2 0.168 - … … … … 15 BCL2- CDK2 0.092 -

7 is a diagram illustrating a method of configuring a functional ontology according to an exemplary embodiment.

Referring to FIGS. 3 and 7, a method of generating a functional ontology from the base ontology of FIG. 6 generated by the functional group indicator discovery apparatus 100 will be described as an example.

For example, assuming that all genes are {MDM2, p53, BAX, CASP8, BAD, PLK3}, functional similarity between genes is calculated as shown in Table 4. According to the functional similarity of Table 4, gene pairs having high functional similarity can be aligned as shown in Table 5.

MDM2 p53 BAX CASP8 BAD PLK3 MDM2 - 0.236 0.190 0.125 0.105 0.163 P53 0.236 - 0.195 0.137 0.124 0.165 BAX 0.190 0.195 - 0.200 0.217 0.168 CASP8 0.125 0.137 0.200 - 0.230 0.103 BAD 0.105 0.165 0.217 0.230 - 0.198 PLK3 0.163 0.124 0.168 0.103 0.198 -

order Gene pair Functional similarity Ontology Node One MDM2-p53 0.236 Base Ontology Node T1 2 BAD-CASP8 0.230 3 BAX- BAD 0.217 4 BAX-CASP8 0.200 Base Ontology Node T2 5 BAD- PLK3 0.198 … …

It describes constructing an ontology using “known function 1: p53 pathway” and “known function 2: apoptosis”. In the same manner as described in FIG. 6, a gene pair having a high functional similarity is added to the gene network one by one, looking for the maximum click, and adding it to the base ontology.

In step 1, the functional group index discovery apparatus 100 generates an empty gene network in which no relationship is formed, and acquires a base ontology.

In step 2, the functional group index discovery apparatus 100 adds the relationship of the gene pair (MDM2-p53) having the highest functional similarity in Table 2 to the gene network, and searches for a maximum click. At this time, the maximum click {MDM2, p53} is a multifunctional gene pair, so it already exists as the ontology node T1. Therefore, {MDM2, p53} does not need to be added to the ontology, so it moves on to the next gene pair. In other words, in the process of constructing a functional ontology, only the maximum click containing at least one gene, not the multifunctional gene, is added to the ontology.

In step 3, BAD-CASP8 is added to the gene network, and a new maximum click {BAD, CASP8} is searched. {BAD, CASP8} contained a gene called BAD, not a multifunctional gene. Also, because {BAD, CASP8} shares “Known Function 1: Apoptosis”, it is added to the ontology as a new end node (T7).

Also in step 4, BAX-BAD is added to the gene network, and a new maximum click {BAX, BAD} is searched. Since {BAX, BAD} shares the known function 1: Apoptosis", it is added to the ontology as a new end node (T8).

In step 5, CASP8-BAX is added to the gene network, and new maximum clicks {BAD, BAX, CASP8} are searched. {BAD, BAX, CASP8} all share “Known Function 1: Apoptosis”, and ontology node T2 (CASP8, BAX), ontology node T7 (BAD, CASP8), and ontology node T8 (BAD, BAX) are {BAD, Since it is a subset of BAX, CASP8}, {BAD, BAX, CASP8} is added to the ontology as the parent node T9 of T2, T7, and T8.

In step 6, PLK3-BAD is added to the gene network, a new maximal click {PLK3, BAD} is not included in the “known function 1: p53 pathway” and “known function 2: apoptosis”, so a shared (common) function. It is judged that it does not have. Therefore, {PLK3, BAD} is not added to the ontology.

In this way, by adding genetic relationships one by one, ontology expansion is performed by adding to the acquired base ontology until all relationships are displayed in the gene network, and the finally obtained ontology can be obtained as a functional ontology. Functional ontology is created with nodes that are expanded from the base ontology node, as shown in Table 6 below. All ontology nodes in the functional ontology can be output as functional group indicators. In this process, relationships between genes are added one by one for explanation, but an ontology can be constructed by adding a relationship set by a specific number or at a specific similarity score interval.

order Gene pair Functional similarity Ontology Node One MDM2-p53 0.236 Base Ontology Node T1 2 BAD-CASP8 0.230 T7 3 BAX- BAD 0.217 T8 4 BAX-CASP8 0.200 T9={T2, T7, T8} 5 BAD- PLK3 0.198 - … … … …

As described above, according to the embodiment, genes having similar functions can be reconstructed into a comprehensive and essential functional group, and through this, it is possible to reflect a disease mechanism complicatedly composed of unit functions within a cell and higher functions in which they are combined. , It can improve the accuracy and stability/reproducibility of disease identification.

According to an embodiment, by integrating and reconfiguring a previously known functional gene set, a gene set with a known causal relationship due to a function can be discovered as one "functional group indicator", and specifically, a "common functional group indicator" related to various functions. The combination of “and “sub-function group indicators” associated with specific sub-functions can select group indicators that can describe disease and cellular functions.

According to the embodiment, it is possible to select a "disease group indicator" showing a specific expression pattern in a disease while having a functional causal relationship through analysis of disease indicators and disease transcript data, and a discrimination model is constructed based on the disease group indicators, The reproducibility of disease state determination can be improved.

The disease group indicator selected according to the embodiment may be widely used for disease diagnosis, prognosis prediction, drug accompanying diagnosis, etc., and may be manufactured and used as a microarray or a multiplex analysis kit.

According to the embodiment, it overcomes the limitations of the existing individual gene expression index, which was inferior in reliability and stability due to difficulty in searching for meaningful mutations in disease, and solves the problem of the existing functional group analysis method that was not able to synthesize and utilize various functional information I can.

The embodiments of the present invention described above are not implemented only through an apparatus and a method, but may be implemented through a program that realizes a function corresponding to the configuration of the embodiment of the present invention or a recording medium in which the program is recorded.

Although the embodiments of the present invention have been described in detail above, the scope of the present invention is not limited thereto, and various modifications and improvements by those skilled in the art using the basic concept of the present invention defined in the following claims are also provided. It belongs to the scope of rights.

Claims (7)

As an operating method of the functional group index excavation device,
Collecting functional groups in which genes with similar functions are grouped, and
Expanding the gene network by sequentially connecting gene pairs based on the similarity of the functional group containing the collected genes, and selecting the gene set of the maximum clique searched in the gene network as a functional group index
Containing, operating method. In claim 1,
The step of selecting as the functional group index
If the gene set of the maximum click searched in the gene network exists in the collected functional groups, the procedure of generating the searched maximum click gene set as a node of the ontology is repeated, and corresponding to each node constituting the ontology. Outputting the gene set to be performed as the functional group index. In paragraph 2,
The step of selecting as the functional group index
If there is a subset node of the gene set corresponding to the searched maximum click in the ontology, a node corresponding to the searched maximum click is added as a parent node of the partial set node,
If there is no subset node of the gene set corresponding to the searched maximum click in the ontology, adding a node corresponding to the searched maximum click as a terminal node,
How it works. In claim 1,
The step of selecting as the functional group index
Extracting multifunctional genes having a frequency of appearance in the functional groups greater than or equal to a reference,
In the order of multifunctional gene pairs with high functional similarity between multifunctional genes, the first gene network is expanded by connecting the multifunctional gene pairs, searching for a maximum click in the first gene network, and a gene set corresponding to the searched maximum click If present in the collected functional groups, generating a base ontology by repeating the procedure of generating the gene set corresponding to the searched maximum click as an ontology node,
Calculate the functional similarity between genes for the collected genes, expand the second gene network by connecting the corresponding gene pairs in the order of gene pairs with high functional similarity, search for maximum clicks in the second gene network, and search If the gene set corresponding to one maximum click exists in the collected functional groups, the function ontology is performed by repeating the procedure of adding the gene set of the maximum click searched in the second gene network to the base ontology based on the base ontology. To generate, and
Outputting nodes constituting the functional ontology as functional group indicators
Containing, operating method. delete delete delete

Images