KR100610240B1 - System and method for reverse-engineering genetic circuits using bayesian network learning - Google Patents

Abstract

The present invention provides a genetic circuit reverse engineering system and method using Bayesian network learning that enables reliable inference of gene interaction even when the experimental data is insufficient compared to the number of genes. The system of the present invention includes a first and second similarity calculating unit for calculating the similarity between genes by using biotin information and mutual information between genes; Seed gene selection unit for selecting a gene having a specificity as a seed gene according to the experimental conditions from the microarray expression data; A gene module forming unit configured to form a gene module that is a similar gene group while expanding a region around the seed gene based on the result obtained through the first and second similarity calculating units; A learning unit that performs parallelized Bayesian network learning on each of the gene modules formed around the gene seeds; It is characterized by consisting of an integration unit for integrating each gene module learned in the learning unit by using a mediator gene commonly included between each of the gene modules into a network.

Gene Interaction, Inference, Bayesian Network, Gene Network

Description

Genetic circuit reverse engineering system and method using Bayesian network learning {SYSTEM AND METHOD FOR REVERSE-ENGINEERING GENETIC CIRCUITS USING BAYESIAN NETWORK LEARNING}

1 is a block diagram of a genetic circuit reverse engineering system using Bayesian network learning according to the present invention.

2 is an overall operational flow diagram of the present invention.

Figure 3 is a view for explaining a method for calculating the similarity between genes using biotin information in the present invention.

4 is a diagram showing a gene module generated through the gene module forming unit of FIG.

5 is an algorithm showing a method of integrating each gene module through a mediated gene in the present invention.

Figure 6 is a diagram showing the intergene interaction network obtained through the present invention.

<Explanation of symbols for the main parts of the drawings>

110, 120: first and second similarity calculation unit 130: seed gene selection unit

140: gene module forming unit 150: learning unit

160: integrated unit

The present invention relates to a gene interaction inference system and method, and in particular, a genetic circuit reverse engineering system using Bayesian network learning that enables reliable interaction inference by minimizing the effects of false inference due to insufficient experimental data compared to the number of genes. And to a method.

The function of cells and their changes are the result of a complex interaction of thousands of genes and their products. These activities can be explained through complex interaction networks that regulate the expression of genes.

Recently, as genetic analysis at the genome unit is possible through microarray experiments, attempts to shed light on the functions and internal mechanisms of living organisms have been actively made by inferring large-scale gene interactions through microarray data.

For this reasoning, several methodologies have been proposed, including Boolean Network and Bayesian Network. Among them, the Bayesian network is accepted as suitable for inference of gene interaction based on theoretical validity and statistical stability.

However, due to the amount of microarray experimental data, which is significantly insufficient compared to the number of genes of thousands, there is a problem of increasing false inference (False Positive).

In addition, the limitation of using Bayesian networks is that it is impossible to deduce the total interactions for thousands of genes.

Conventionally, there is a small candidate method for reducing the range of interaction inference or a model averaging method for extracting a common one from a plurality of inferred models (Pacific Symposium on Biocomputing, Hartemink et al. 7, pp. 437-449, 2002), attempts have been made to deduce from known biological information, but the lack of fundamental data has not made much progress.

The present invention has been made to solve this problem, and an object of the present invention is to reverse the genetic circuit engineering system using Bayesian network learning to minimize the false inference generated from the lack of experimental data and to build a more reliable gene interaction network and In providing a method.

Another object of the present invention is to provide a genetic circuit reverse engineering system and method using Bayesian network learning capable of inferring the entire genome-wide gene.

Genetic circuit reverse engineering system using Bayesian network learning according to the present invention for achieving the object of the present invention, the first, second similarity calculation means for calculating the similarity between genes using biotin information and mutual information between genes ; Seed gene selection means for selecting a gene having a specificity as a seed gene according to experimental conditions from the microarray expression data; A gene module forming means for forming a gene module which is a similar gene group while expanding a region around the seed gene based on the result of the first and second similarity calculating means; Learning means for performing parallelized Bayesian network learning on each of the gene modules formed around the gene seeds; And integrating means for integrating each gene module learned in the learning means into one network by using a mediator gene commonly included between the respective gene modules.

Genetic circuit reverse engineering method using the Bayesian network learning of the present invention for achieving the above object, the first step of calculating the similarity between each gene using biotin information and inter-gene mutual information; Selecting a gene having a specificity as a seed gene according to experimental conditions from the microarray expression data; A third step of forming a gene module which is a similar gene group by expanding a region centered on the seed gene based on the result obtained through the first step; Performing a parallelized Bayesian network learning on each of the gene modules formed around the gene seeds; And a fifth step of integrating the learned genetic modules into one network by using a mediator gene commonly included between the respective genetic modules.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings. However, the following examples are merely to illustrate the present invention is not limited to the contents of the present invention.

1 is a block diagram of a genetic circuit reverse engineering system using Bayesian network learning according to the present invention.

As shown, a first similarity calculation unit 110 for calculating similarity between genes using biological annotation information, and a method for calculating similarity between genes using mutual information between genes. 2 similarity calculation unit 120, seed gene selection unit 130 for selecting a gene having a specificity according to experimental conditions (Experimental Condition) from the microarray expression data (Sear Gene) (Seed Gene), 130, On the basis of the results obtained through the similarity calculation unit 110 and 120, the genetic module that forms a gene module (module 1-module n), which is a similar gene group, is expanded while the region centered on the seed gene. Forming unit 140, the learning unit for performing parallelized Bayesian network learning for each gene module (module 1-module n) made around the gene seeds (15) 0), each of the gene modules (module 1-module n) learned in the learning unit 150 by using an intermediary gene (Communication Gene) commonly included between each of the module (module 1-module n). It is composed of an integrated unit 160 to integrate into one network.

The present invention configured as described above will be described with the flowchart of FIG. 2.

First, the first similarity calculation unit 110 for measuring similarity between genes using the biotin information quantifies the similarity of functions between genes through a score (S110).

As shown in FIG. 3, when the functions of the two genes are different, the two common parent nodes are found in the tree structure and their values are taken. Similarity is obtained by applying a common RSS (Resnik Sementic Similarity) to this value.

Looking more closely at Equation 1, if S (f _i , f _j ) is a negative log of the values of the shortest common parent nodes of the two nodes, then the similarity AI (g _i , g _j ) between the two genes is common. It is obtained by adding all the S (f _i , f _j ) values in the case of having a function and adding the maximum S (f _i , f _j ) values for different functions.

수학식 1

Equation 1

In addition, the second similarity calculator 120 that calculates similarity between genes by using mutual information between genes may quantify similarity between genes based on expression patterns of genes (S110). For this purpose, mutual information between genes is used as a measure.

That is, when g _i and g _j are genes and x _i and y _j are discretized expression values of the genes g _i and g _j , MI (g _i , g _j ) is obtained by the following equation.

Equation 2

MI (g _i , g _j ) = ∑x _i ∑x _j p (x _i , x _j ) log (p (x _i , x _j ) / p (x _i ) p (x _j ))

In Equation 2, P is a probability function.

In addition, the seed gene selection unit 130 extracts a gene representing each experimental condition based on the gene expression pattern (S120).

That is, u _ci is defined as the mean of gene i under experimental condition c, u _￢ci is the mean of gene i under conditions except experimental condition c, and the standard deviation in each is defined as σ _ci , σ _￢ci . At this time, among the specificity measure D values obtained as in Equation 3 below, genes of a threshold u _D + 3 * σ _D or more are selected as seed genes. Where u _D and σ _D are the mean and standard deviation of the specificity measures, respectively.

Equation 3

D = | u _ci - u _￢ci │ / (σ _ci + σ _￢ci )

In addition, the gene module forming unit 140 forms a gene module (module 1-module n) as shown in FIG. 4 while expanding a region around the seed gene as a union of biotin information and mutual information (S130).

For this expansion, the relevant genes are selected using previously known biotin information from experiments, while mutual information based on gene expression patterns is used to include genes that are not previously known.

As reference values for region expansion for forming the module 1-module n, u _AI + 4 * σ _{AI and} u _MI + 3 * σ _MI are used, respectively. Where u _AI is the mean of similarity between genes using biotin information, σ _AI is the standard deviation of similarity between genes using biotin information, u _MI is the mean of similarity between genes using mutual information, and σ _MI is the standard deviation of similarity between genes using mutual information.

In addition, the reference value is to adjust the size of the gene module (module 1-module n) to be formed, and to balance the amount of genes introduced from mutual information and bio tin information, it is an empirical value through experiments.

The learning unit 150 maximizes the Bayesian network learning speed through parallel processing for each gene module (module 1-module n) made based on the seed gene, and for this purpose, each gene module based on the seed gene. (module 1-module n) is divided into threads (Thread) unit, the learning proceeds (S140).

Hill Climbing, Sparse Candidate, and Model Averaging are used for effective learning, and Minimal Description Length Score is used to evaluate the network structure. .

In addition, since genes are inferred by each gene module (module 1-module n) without inferring thousands of genes at once, the ratio of experimental data per gene is greatly improved, which greatly reduces the possibility of inference error.

In addition, the integration unit 160 is to make each genetic module (module 1-module n) centered on the seed gene to be an integrated organism without remaining as a separate network result separated from each other (S150). Conventional methods use clustering to create several groups, but have no link between them, but only separate groups.

Therefore, in the present invention, the entire gene module (module 1-module n) is integrated into one network by using a mediator gene commonly included between each gene module (module 1-module n). Fig. 5 shows the algorithm of this work, and finally the gene interaction network as shown in Fig. 6 is obtained.

As described above, although described with reference to a preferred embodiment of the present invention, those skilled in the art various modifications of the present invention without departing from the spirit and scope of the invention described in the claims below Or it may be modified.

As described above, the present invention can construct a reliable gene interaction network by minimizing the influence of false inference due to insufficient microarray experimental data, and also enables the analysis of large genes at the genome level. You get a comprehensive view as well as a detailed view of the action.

Claims (8)

 First similarity calculating means for calculating similarity between genes using biotin information; Second similarity calculating means for calculating similarity between genes using mutual information between genes; Seed gene selection means for selecting a gene having a specificity as a seed gene according to experimental conditions from the microarray expression data; A gene module forming means for forming a gene module which is a similar gene group while expanding a region around the seed gene based on the result of the first and second similarity calculating means; Learning means for performing parallelized Bayesian network learning on each of the gene modules formed around the gene seeds; And Integration means for integrating each gene module learned by the learning means into one network by using a mediator gene commonly included between the respective gene modules; Genetic circuit reverse engineering system using Bayesian network learning, characterized in that consisting of. The method of claim 1, wherein the first similarity calculating means Calculate similarity between genes using RSS (Resnik Sementic Similarity), but when S ( f _i , f _j ) takes a negative log at the value of the shortest common parent node of two nodes, the similarity AI ( g _i) between the two genes , g _j ) is a Bayesian which is obtained by adding all S ( f _i , f _j ) values for common functions and adding the maximum S ( f _i , f _j ) values for different functions. Genetic Reverse Engineering System Using Network Learning. The method of claim 1, wherein the second similarity calculating means g _i and g _j are genes, x _i and y _j are discretized expression levels of the genes g _i and g _j , and P is a probability function, where MI (g _i , g _j ) is the similarity between genes. Genetic reverse engineering using Bayesian network learning, characterized by _i ∑x _j p (x _i , x _j ) log (p (x _i , x _j ) / p (x _i ) p (x _j )) system. According to claim 1, wherein the seed gene selection means When u _ci is defined as the mean of gene i under experimental condition c, u _￢ci is the mean of gene i under conditions except experimental condition c, and the standard deviation in each is defined as σ _ci , σ _￢ci . D = | u _ci - _Genes greater than or equal to the reference value u _D + 3 * σ _D (u _D : mean of specificity measures, σ _D : standard deviation of specificity measures) among the specificity measures D obtained as u _￢ci │ / (σ _ci + σ _￢ci ) Genetic circuit reverse engineering system using Bayesian network learning, characterized in that the selection as a seed gene. The method of claim 1, wherein the gene module forming means Genetic modules based on seed genes are formed by combining the biotin information and mutual information, wherein u _AI is the mean of similarity between genes using biotin information, and σ _AI is the standard deviation of similarity between genes using biotin information. , When u _MI is the mean of similarity between genes using mutual information, and σ _MI is the standard deviation of the similarity between genes using mutual information, u _AI + 4 * σ _{AI and} u _MI + are the reference values for forming the genetic module. Genetic circuit reverse engineering system using Bayesian network learning, characterized by using 3 * σ _MI . The method of claim 1, wherein the learning means Genetic circuit reverse engineering system using Bayesian network learning, characterized by parallel processing of each gene module in units of threads. The method of claim 1, wherein the learning means A genetic circuit reverse engineering system using Bayesian network learning, which uses a hill climbing method or a small candidate method or a model merging method for learning, but uses a Minimum Description Length Score (MDL) score as a measure of network structure evaluation. A first step of calculating similarity between genes using biotin information and mutual information between genes; Selecting a gene having a specificity as a seed gene according to experimental conditions from the microarray expression data;  A third step of forming a gene module which is a similar gene group by expanding a region centered on the seed gene based on the result obtained through the first step; Performing a parallelized Bayesian network learning on each of the gene modules formed around the gene seeds; And A fifth step of integrating the learned genetic modules into one network by using a mediator gene commonly included between the respective genetic modules; Genetic circuit reverse engineering method using Bayesian network learning, comprising a.

Images