KR101003175B1 - The method to identify the multipurpose potential gene using cross-talk mapping - Google Patents

Abstract

The present invention relates to a method for discovering genes involved in regulating cell function from microarray data. Specifically, crosstalk genes are selected through a process of effectively combining previously developed data analysis methods. It is about how to. The method of the present invention not only found a large number of genes associated with the response to abiotic stress stimulation of the Arabidopsis, a representative model plant, but also screened for response specific genes and correlation genes for the response.

Microarrays, Gene Mass Discovery, Data Mining, Graph Visualization, Reverse Engineering, Intracellular Networks

Description

The method to identify the multipurpose potential gene using cross-talk mapping}

The present invention relates to a method of discovering genes involved in regulating cell function from microarray data.

Microarray is a technique that is mainly used to observe the expression changes of these genes simultaneously in an environment where a large number of gene mutations are expected, such as cancer or stress stimulation. The technique has the advantage of simultaneously analyzing the expression changes of thousands of genes in cells for internal / external stimulation, but analyzing the large amounts of numerical data generated from the analysis results and linking the results with biological phenomena. I have a hard time to do it. To analyze gene expression profiles obtained using microarray technology, many analytical methods of numerical analysis and many tools for implementing them have been presented. Most of these genes are analyzed by comparing and analyzing the relative expression differences of genes, selecting genes related to the stimulus, and predicting the function of newly selected genes with them based on the functions of known genes. It has been used to look at the biological reaction mechanisms. In recent years, more sophisticated data analysis methods have evolved to infer the interrelationships between these genes. There are several analysis tools that I find very useful for microarray data analysis.

First and foremost is VxInsight, a data-mining software developed by Sandia National Laboratories (www.sandia.gov) in the United States. VxInsight makes it easy to discover the information implied by the data by presenting a two-dimensional or three-dimensional picture of the interrelationships between the large amounts of data stored in the database. The tool was already used in 2001 for the microarray data analysis of Caenorhabditis elegans at the Stuart Kim Lab at Stanford University to select co-expressed genes (SCIENCE, VOL 293, 2001, 2087 -2092). The Arabidopsis Information Resource (www.arabidopsis.org) also uses this tool to analyze microarray data from large projects such as the Arabidopsis Functional Genomics Consortium (AFGC) and the AtGenExpress Consortium. It provides related information.

VxInsight is divided into two parts, one called the Ordination routine, called VxOrd, that calculates their layout on the two-dimensional plane using similarities between data objects. (Figure 1), the other is a visualization engine (visualization engine) for visualizing the x, y coordinates of the calculated gene. Since VxOrd calculates that genes with similar expression patterns are located close to each other based on the similarity of all the gene pairs present in the microarray expression profile to be analyzed, the two-dimensional coordinate values of genes expressed together are concentrated in very similar regions. Show a tendency. This means that many genes are expressed together in many cases, and VxInsight displays these clustering results in not only two-dimensional but three-dimensional mountain maps to mount the number of genes that are expressed together. Reflect as the height of. That is, if the number of genes concentrated in the position A is greater than the number of genes concentrated in the position B, the height of the mountain formed in the region A becomes higher than B. VxInsight's visualization engine is a tool that makes it easy to grasp the information implied by a large amount of data.However, in the case of genes having a long continuous sequence, the genes at both ends have a similar expression pattern. As it is difficult to see the data, the criteria for separating them into sub clusters are needed again when interpreting the data.

Second, we classify genes based on known information, such as biological function or physical location on the genome, and use gene expression to group genes that belong to the same category into a group of genes called a gene set. There is a statistical method called Gene Set Enrichment Aanlysis that analyzes the data to determine which groups of genes are involved in representing phenotype differences. Developed by the Broad group of the MIT Institute of Technology in the early 2000s (Nature genetics 2003, vol 34 267-273, PNAS 2005, 102, 15545-15550), for this method the function or genome of the gene product In addition to the classification according to the location of the images, if the genes are organized in various aspects by classifying the genes with various criteria, it is possible to obtain a great help for a deeper understanding of life phenomena by which gene bundles are activated in which situation. However, if the gene bundle, which is the core of the analysis, needs to be preceded, and the number of gene bundles and groups to be analyzed is increased, it is again time-consuming to understand the biological mechanisms by integrating the results obtained. .

Finally, `` ARACNe (Algorithm for the Reverse Engineering of Accurate Cellular Networks) '' (nature genetics vol37, 382-), which reconstructs cellular networks from microarray data developed by the California College of Columbia, USA. And reverse engineering algorithm (390). This algorithm calculates a mutual information scale that represents the correlation between genes. Mutual information (MI) is referred to here as a probability theory or information theory. Compare the degree to which A appears, the degree to B, and the degree to which AB appears together. If A or B are always AB together, there will be an inevitable relationship between A and B. It is the information on which the judgment is made. Therefore, this algorithm helps to better understand the interaction phenomenon between genes by finding gene pairs with MI values above a certain level of confidence. However, in order to proceed with the analysis using this algorithm, at least 100 pieces of microarray data are required in principle due to the reliability of the results.

While all of the assays discussed so far help us to understand the pre-reactor behavior of cells from gene expression data, they do not provide a way to meet the conditions that must be premised on maximizing these assays. As mentioned above, in the case of the GSEA method, the analysis of data requires the construction of a very precise and high-volume gene bundle, and even when the reverse engineering algorithm is implemented, a large number of microarray data are generated due to reliability problems. need. In spite of the fact that various methods of analysis of microarray data have been proposed to date, many researchers still incorporate numerical data from gene expression profiles into gene-level understanding in understanding cellular responses to internal and external stimuli. This problem is thought to be one cause of the great difficulty in making this work.

Therefore, the present inventors feel the necessity of developing a method of extracting biologically useful information, and appropriately modify and supplement the aforementioned three microarray data analysis methods, and combine them to induce synergistic effects, and then function the cells from the gene profile. We have developed a method to quickly discover the relationship between a group of genes involved in regulating the expression of proteins and external stimuli.

In addition, when this is applied to plants in which it is difficult to obtain a large amount of microarray data tested under similar conditions, for example, when analyzing expression profiles of Arabidopsis genes exposed to environmental stress, a large amount of Arabidopsis involved in environmental stress response. The present invention was completed by showing that not only the discovery of genes but also the interaction relations between these genes can be easily observed through comparison with known literature information.

The combination of micro data analysis methods provides a method of discovering a large amount of genes that are commonly expressed in specific conditions such as stress, and analyzes the interaction relationship of these genes to select response-specific or multi-stimulatory genes. To provide a way.

In addition, by combining the computational unit associated with the micro-data analysis method to discover a large number of genes commonly expressed in a specific condition, such as stress, and to analyze the interaction relationship of these genes to select a response-specific or multi-stimulation genes To provide.

In order to achieve the above object, the present invention uses gene bundles to group gene groups that exhibit similar expression patterns based on clustering results of a large amount of microarray data analyzed by a data mining program. By statistical analysis, it provides a method for rapid mass discovery of gene bundles expressed under specific conditions.

In addition, the present invention, in addition to the above method, provides a method for screening genes that are correlated by applying a reverse engineering program to the genes of the discovered gene bundle.

In addition, the present invention provides a device from which microarray data are discovered, in which a large number of genes involved in regulating the function of a cell are discovered.

The present invention also provides an apparatus for selecting correlated genes, including an operation unit for reconfiguring an intracellular network in addition to the above method.

The present invention effectively combines and processes the methods used for microarray data analysis to discover large quantities of response genes and to select genes that are specific or cross-talk specific to a specific reaction or stress. Make it happen.

In addition, the present invention introduces an overlap concept to cluster continuous data that is difficult to divide, thereby avoiding an opportunity deprivation situation in which an individual belongs to one cluster and does not belong to another type of cluster.

In addition, the present invention analyzes the interaction between the highly significant genes selected through statistical analysis method by applying reverse engineering algorithms to overcome the limitations of the conventional method, which requires a large amount of microarray data as a reliability problem. To overcome.

In addition, according to the present invention, selection of gene groups associated with specific genes and their interactions can be easily understood.

Due to the analytical method according to the present invention, it is possible to easily derive a hypothesis explaining the reaction of cells to an external stimulus at the gene level and help design a verification experiment therefor.

In order to solve the above problems, the present invention

A first step of inputting a gene obtained on the basis of gene expression similarity in the microarray data and a relative position in which the gene is expressed in two-dimensional coordinates by using a data mining program for determining the relative position of genes;

A second step of determining a boundary size for classifying the two-dimensional coordinates of the gene input in the first step into a gene bundle;

A third step of creating a new gene bundle catalog having four different origins based on the boundary size of the gene bundle determined in step 2;

Statistical analysis is applied to the gene bundle catalog created in the third step, microarray data is obtained for normal cells (control) and stimulated cells (experimental group), and then the microarray data is compared and analyzed for genes. A fourth step of displaying an expression aspect;

Based on the analyzed data of the fourth step, using a visualization program loaded into the microprocessor and the memory of the computer, the data associated with all the analyzed gene bundles of the fourth step are visualized to select the correlated gene bundles. It provides a method of discovering a large amount of genes involved in regulating the function of the cells in the microarray data comprising a fifth step.

In this case, the microarray data used for constructing the gene bundle and for comparative analysis is experimental data using Affymetrix ATH1 (25K) arrays stored in a microarray database provided by TAIR (The Arabidopsis Information Resource, //www.arabidopsis.org). This is not limited to gene expression omnibus-// www.ncbi.nlm.nih.gov/geo/ of the NCBI or array express (// www.ebi.ac.uk/microarray -as / ae / of the EBI. Experimental data using the various types of microarrays provided by) can be used in the same way to construct coexpressed gene bundles in organisms other than Arabidopsis.

In addition, the microarray data may also use microarray data obtained in a laboratory to discover genes commonly expressed between specific stimuli of a specific organism.

In addition, the mining program is a data analysis method for analyzing correlations and patterns in a large amount of microarray data. In the present invention, the mining program is based on VxOs of VxInsight provided by Sandia National Lab. (//Www.sandia.gov). However, a program called STN AnaVist (//www.stn-international.de/stninterfaces/stnanavist/stn_anavist.html) can also be used, as long as you can obtain analytical data on associations and expression patterns in the microarray. It is not limited.

In addition, since the relative position data reflecting the expression similarity of the genes calculated in the first step belongs to continuous data in nature, another criterion is needed to separate them into sub-clusters when interpreting the data. To solve this problem, the inventors created four types of gene bundle catalogs with different origins. In other words, by changing the starting point of the rectangle of length 15, which is a standard for separating genes displayed on a two-dimensional plane into a bundle unit, (1) UL (Up-Left) 2) UR (Up-Right) 3) DL (Down- Left) 4) DR (Down-right)] Four types of gene bundle catalogs were prepared.

The boundary size of the second stage means the distance between genes that are recognized as co-expression and can be recognized as one co-expressed gene bundle, and is determined to include 4 to 20 genes in the single gene bundle. It is preferred, but not limited thereto. Accordingly, the size of the boundary may be 10 to 20, and preferably 15, but is not limited thereto.

In the case of dividing into four types of origin as described above, the data is integrated through the overlap during visualization after processing GSEA statistics. The introduction of the overlap concept clusters continuous data that is difficult to partition, thus avoiding the opportunity deprivation situation where an entity belongs to one cluster and does not belong to another. In addition, since it is calculated as a cluster, it is possible to significantly reduce the computational burden, thus enabling an efficient analysis method.

In addition, the statistical analysis method of the fourth step is preferably a gene set enrichment analysis (GSEA) method developed by the broad group of the MIT University of Technology, but a method called parametric analysis of gene set enrichment (PAGE) (Ruan). J. et al., BMC Bioinformatics. 2005; 6: 114) and the like are also possible and are not limited to this as long as similar calculations are made.

In addition, the visualization program is a program for visualizing the data obtained through the four steps. Using these programs, you can immediately find key genes. Available programs include graph visualization (Graphviz) (http://www.graphviz.org/), Cytoscape (//www.cytoscape.org/) developed by the American Institute for Systems Biology, or Vladimir Batagelj And Pajek (//vlado.fmf.uni-lj.si/pub/networks/pajek/) co-developed by Andrej Mrvar, but the 'Graphviz' is used in a specific embodiment of the present invention, but is not limited thereto. .

Genes belonging to the gene bundle selected in the fifth step can be determined genes involved in regulating the function of the cell in response to stress and stimulation.

That is, through the method of the present invention it is possible to discover a large amount of genes involved in regulating the function of the cell, and through a visualization program it is possible to quickly and easily find specific or correlated gene bundles and genes.

In addition, the present invention is divided into the coherence and contribution to the distribution of the core gene in the bundle according to the stimulus for the gene bundle selected in step 5 in addition to the above step, these gene bundles according to the related degree of response to the stimulus Iii) a bundle of truly correlated genes with at least 0.5 cohesion and a contribution that correlates with all stimuli; ii) a bundle of genes with a cohesion of less than 0.5 but a contribution of at least 0.5; Bundle iii) a sixth step of finding a real cross-talk gene bundle after four subdivisions, such as a false positive gene bundle with both cohesion and contribution less than 0.5; And

Reconstruct intracellular networks using a program that implements a reverse algorithm loaded in a microprocessor and memory for genes included in the real cross-talk gene bundle found in step 6 above. The seventh step of selecting genes that are predicted to be relevant

It further provides a method for screening genes that are interrelated.

In this case, 'contribution' refers to the percentage of the core genes that played a decisive role in winning the bundle for the total number of genes selected in a specific stimulus. As the value approaches 1, many genes in the bundle respond to the stimulus.

In addition, 'aggregation' refers to the ratio of the number of genes analyzed as key genes in all A, B, and C stimulus in the total number of genes when a bundle of genes wins a bundle of genes in response to multi-stimulation. The closer this value is to 1, the more likely that these genes are not only gene bundles that are well bound as a functional unit, but are also likely to be biologically useful real cross-talk gene bundles (FIG. 13).

In this case, the reverse engineering program used is a program that helps to predict the composition of the network between genes by calculating a mutual information scale value representing the correlation between genes. In the present invention, 'ARACNe' (nature genetics vol 37, 382-390) is used, but is not limited thereto. Linear Model (BMC Bioinformatics. 2004 10; 5: 108), Bayesian (BMC Syst Biol. 2008 4; 2: 57 ) Or Gaussian network model (Genome Res. 2007 17 (11): 1614-25) are also applicable. Using this program, it was possible to confirm which mechanisms of the 'interrelated genes' selected in the present invention were actually included in the intracellular network.

The present invention utilizes a number of microarray results known in the art and analyzes expression patterns thereof through a combination of the above-described analytical methods, thereby mass-producing reaction genes that interact with specific stimuli, and at the gene level. Observing the interaction provides a way to select response specific genes.

In addition, the present invention provides an apparatus for discovering a large amount of genes involved in regulating the function of cells in order to achieve the above object. The apparatus is composed of three parts: an input unit for inputting data, an operation unit for processing the input data, and an output unit for obtaining the processed result. The device is described in more detail as follows.

It provides an input unit for receiving the relative position data of the gene expressed on the basis of the expression similarity of the gene in the microarray data and the gene represented by the two-dimensional coordinates of the gene.

In this case, the microarray data used for constructing the gene bundle and for comparative analysis is experimental data using Affymetrix ATH1 (25K) arrays stored in a microarray database provided by TAIR (The Arabidopsis Information Resource, //www.arabidopsis.org). This is not limited to gene expression omnibus-// www.ncbi.nlm.nih.gov/geo/ of the NCBI or array express (// www.ebi.ac.uk/microarray -as / ae / of the EBI. Experimental data using the various types of microarrays provided by) can be used. In addition, various microarray data obtained in the laboratory can be used.

In addition, the operation of calculating the coordinate value to be located on the two-dimensional plane by reflecting the expression similarity of the genes shown in the data is based on VxInsight provided by Sandia National Lab. (//Www.sandia.gov) in the present invention. Although it is used, a program called STN AnaVist (// www.shinwon.co.kr/cas/products/anavi st / index.html) is also available, and it can provide location data by analyzing the association and expression patterns in the microarray. The type is not limited as far as possible.

In addition, the following calculation unit is provided,

A first calculating unit classifying the two-dimensional coordinates of the input gene into gene bundles according to a predetermined boundary size;

A second operation unit classifying a new gene bundle catalog having four different origins based on the boundary size of the gene bundle;

A third calculation unit which statistically analyzes the gene bundle catalog to obtain microarray data of normal cells (control) and stimulated cells (experimental group), and then compares the values to display expression patterns of genes;

A fourth operation unit which visualizes the gene bundle expression map and selects the correlated gene bundles by using a visualization program loaded in a microprocessor and a memory of a computer based on the data obtained from the three calculation units; And

It provides a device for discovering a large amount of genes involved in regulating the function of the cell consisting of an output means for outputting the operation result.

In this case, since the relative position data reflecting the expression similarity of the genes obtained by the first calculation unit belongs to continuous data in nature, another criterion is required to separate them into lower clusters when interpreting the data. To solve this problem, the present inventors constructed coordinates having different origins. In other words, the gene bundle obtained from the operation unit 2 is 1) UL (Up-Left) 2) UR (Up-Right) 3) DL (Down-Left) 4) DR (Down-right) Create a gene pool catalog of types.

In addition, the statistical analysis method of the fourth operation unit is preferably a method of gene set enrichment analysis developed by the broad group of the MIT University of Technology, and a method called parametric analysis of gene set enrichment (PAGE) (Ruan J et al., BMC Bioinformatics. 2005; 6: 114) and the like are also possible, but are not limited thereto.

In addition, the present invention, in addition to the fourth operation unit of the operation unit of the excavation device of a large amount of genes involved in regulating the function of the cell,

Based on cohesion and contribution to gene bundles responding to multistimulation ⅰ) Genuinely correlated gene bundles with cohesion and contribution of at least 0.5 that correlate to all stimuli A coherent gene bundle showing cohesive force of 0.5 or more but less than 0.5 iii) a fifth calculating unit for classifying a coherent gene bundle after classifying into a false positive gene bundle having both cohesiveness and contribution of less than 0.5;

A sixth operation unit configured to reconstruct an intracellular network for genes included in a truly correlated gene bundle found by the fifth operation unit, and to select genes that are predicted to be correlated intracellularly;

It provides a screening device for the correlation of genes, characterized in that it further comprises an output means for outputting the operation result.

In this case, the reverse engineering program used is a program that helps to predict the composition of the network between genes by calculating a mutual information scale value representing the correlation between genes. 'ARACNe' (nature genetics vol 37, 382-390), Linear Model, Bayesian or Gaussian networks are available. ARACNe is preferred but not limited thereto.

The present invention utilizes a number of microarray results that are known in the past through a combination of computational units and analyzes expression profiles thereof, thereby mass-producing reaction genes that interact with specific stimuli, and interact at the gene level. The present invention provides a device capable of selecting reaction-specific genes.

Hereinafter, the present invention will be described in detail with reference to examples.

However, the following examples are intended to specifically illustrate the present invention, and the content of the present invention is not limited by the following examples.

< Example  1> Composition of coexpressed gene bundles of Arabidopsis genes

Of the microarray data stored in TAIR's microarray database using VxOrd from VxInsight, a data-mining software developed by Sandia National Laboratories (www.sandia.gov) Location data on their XY plane reflecting the expression-likeness of more than 28,000 Arabidopsis genes shown in 1436 microarray data sets using the Affymetrix ATH1 (25K) genome arrays as a platform Downloaded and decided to create a co-expression gene bundle as the basic data (Fig. 2). However, there are a few things to consider when constructing gene bundles using the relative two-dimensional positional data of the genes obtained here. First, it is to define the distance between genes that can be recognized as co-expression, that is, how close the genes are to be members of the same gene bundle. Secondly, there is a position value that can be recognized as co-expression, but there should be no genes that do not belong to one gene group member. This occurs because the position values of genes are continuous. In other words, although the gene 'a1' is close to the genes' a2 and 'a3', it is a member of the gene bundle A, but it is also close to the gene 'b1' and co-expressed. However, by standard 'a1' belongs to the 'A' group and 'b1' belongs to the 'B' group. Finally, the false discovery rate means the rate at which false positives are mixed among the genes within the gene bundle that are determined to be statistically significant. Because (FDR) values are influenced by the entire gene bundle involved in the analysis, redundancy between gene bundles within a catalog should be avoided as much as possible. Generating a gene bundle so that it does not miss all possible groups of gene bundles results in little difference between gene bundle members, which can affect GSEA (gene bundle fidelity analysis) results.

Considering the above three points, the co-expression gene bundle can be recognized as co-expression. The distance between genes can be recognized as 10 to 20. 15 which can be satisfied is preferable. In the present invention, the boundary size is defined as 15 so that the size of the rectangular frame that can be defined as one gene bundle is the same.

In other words, the x / y coordinates are divided into blocks of small tetrahedrons with a distance of 15, and the genes located in these blocks are grouped into the same gene bundle. In order to solve the problem of not belonging to a different gene bundle and a group even though it is a short distance, four types of (UL, UR, DL, DR) gene bundle catalogs having different origins were constructed when constructing this tetrahedral block ( 3). That is, if the XY coordinates are divided into blocks having a boundary size of 15, the gene bundle catalog can be created with the origin coordinates as (0,0), (7.5,0), (0,7.5), (7.5,7.5). . The gene bundle catalogs with different origins are then used for analysis as a unit.

 When analyzing the microarray data using the GSEA method, these gene catalog blocks were used as different gene bundles to solve the aforementioned two problems, the combination of overlapping and missing gene bundles, as well as the interpretation of the results. By using these four catalogs, we have the advantage of finding more accurate co-expressed gene bundles by tying up the gene expression gene bundles. As mentioned earlier, the two-dimensional coordinate values of the genes calculated by VxInsight represent the relative relationship of these genes based on the changes in Arabidopsis gene expressions in 1436 gene expression profiles using ATH1 (25K) microarrays. Can be. Therefore, as described above, the co-expression gene bundle prepared by applying this has the advantage of making predictions about gene modules used by living organisms in response to specific stimuli or signals.

< Example  2> GSEA Analysis of gene bundles and visualization of results

<2-1> Co-expression Gene Bundle GSEA  analysis

To enhance the molecular biological understanding of how the limited genetic resources of plants operate and respond to customized responses, the co-expressed gene bundles and GSEA methods previously constructed were used to analyze microarray experimental data associated with environmental stress. The microarray data downloaded from TAIR for analysis was used to treat nine abiotic stresses on the roots and stems of the Arabidopsis, 18 days after sowing, using the same microarray called Affymetrix ATH1. By performing microarray experiments on, time series data designed to view gene expression changes over time (FIG. 4). We conducted GSEA using 30-minute, 1-hour, and 3-hour data to compare the initial response by stress among the 6 independent time-series data for the roots and stems. The genes were then marked with increasing or decreasing expression levels in each condition (FIG. 5). As a result, the GSEA method was used to select gene bundles that were shown to be associated with early response to Arabidopsis depending on the stress environment, and each stress In the treatment, when the number of gene bundles showing a contrasting expression difference was divided into roots and stems, it was found that the change in early gene expression of the Arabidopsis to stress differed not only by stress but also by tissue. For example, comparisons between stresses indicate that the initial response to cold stress involves a large number of gene bundles, unlike salt or heat stress. It was found that the bundles initially increased expression.

<2-2> GSEA Coexpressed Gene Analysis and Visualization

The GSEA results of the cohort of gene expression data used in Arabidopsis early gene expression data analysis for nine types of abiotic stresses are shown in the graph above. It can also be viewed as multidimensional data. Therefore, it is very important that these results be summarized and compared so that they can be meaningful for understanding the plant's initial response. However, the type of location of the different origins and their names (e.g. 23x34y), represented by the location coordinates, are intuitive for changing the overall Arabidopsis gene's pattern of internal and external stimuli, even after basic cleanup. To make the understanding difficult, the inventors have developed a method of visualizing it for easier interpretation of the results.

Using the two-dimensional coordinate values of genes calculated with VxOrd, we can simply draw a two-dimensional graph. We apply this to the x / y positional coordinate value indicated by the gene bundle name and the starting point of the type code. By combining the overlapped co-expression map (ovelapped co-expression map) and marking it in the position corresponding to the selected gene (enriched gene set), it was possible to easily identify the expression changes of the entire gene bundle (Fig. 6). ).

FIG. 7 shows the gene bundles of Arabidopsis genes expressed at 30 minutes / 1 hour / 3 hours when the initial reaction is expected to occur after stress treatment in each stress situation using the co-expression map. In this figure, blue is shown to be activated in the A-group and red is shown in the B-group, which shows that different gene bundles are activated differently according to stress and the overall trend of the distribution of co-expressed gene bundles is shown. The approximate degree of similarity of reactions can be estimated through. For example, in the case of stress due to ultraviolet rays and physical wounds, it is easy to grasp the similarity of the entire pattern, and it can be easily seen through only the map that gene bundles showing high fidelity are very similar (FIG. 7).

< Example  3> GSEA  Identification of stress specificity or general stress gene bundles by integration of assay results

To compare the gene bundles that are activated in each stress environment through GSEA in more detail, the names of genes that express many types of stress that cross the results in a matrix form as shown in FIG. The table is summarized. This table was used to determine whether the gene bundles were stress-specific or common to stress. In particular, in the case of a gene bundle that is commonly activated in various stresses, a bundle of genes related to cross-talk between stress signaling pathways is likely to be gathered, so it will be referred to as a correlated gene bundle. . In the case of these correlated gene bundles, there is a gene bundle in which expression is increased in only one of the A-group or B-group depending on the association with the stimulus, or the gene bundle in which the expressed group varies depending on the stress environment. there was.

As the GSEA for each stress-related gene expression profile was assigned to the A-group and the data for the experimental group to the B-group, it was indicated that a gene group increased expression in the B-group. The expression of more than a certain number of genes in the bundle increased compared to the normal Arabidopsis, meaning that if the expression increased in the A-group, the expression decreased. When the root tissue is cold, genotoxic, heat, oxidative, salt, or wound stress, one of the gene bundles activated in the A-group 'DL_38x19y' and one of a group of genes that increase expression in the B-group when subjected to drought, genotoxic, osmotic, salt, or UVB stress A tree-view visualization program developed by UC Berkeley's Eisen lab (//rana.lbl.gov) for time-domain expression profiles corresponding to the initial (30min, 1h, 3h) genes of DL_55x55y '. By using a diagram showing the image of the signal intensity on the gene expression profile (Fig. 9), the expression of the genes with increased expression in the A-group is mostly reduced compared to the normal state, and conversely, the B-group The expression of genes that are enriched in You can easily see that it is increased. At this time, the X-axis in Figure 8 means the experimental conditions, that is, 30 minutes, 1 hour, 3 hours. The Y-axis represents one gene, and the ratio of gene expression in steady state and stress condition is indicated. If more gene expression is normal, the green color is increased, and the opposite is dark red color.

Through the above visualization program, it is possible to easily discover the general bundle of stress genes that are stress-specific or expressed in all stresses.

< Example 4> using visualization program By condition GSEA  Consolidation of results

The matrix table mentioned above makes it easy to determine whether individual gene bundles are related to nine different stress environments, but nevertheless there is a cross between the stresses and the entire activated gene bundle. There is a difficulty in comprehensively understanding the relationship between the talk or the stress specificity. In order to solve this problem, a graph visualization program called 'Graphviz' is used to group bundles of genes whose expression is enhanced in whole, only in the direction of increasing expression, in the direction of decreasing, or increasing with stress. After dividing into three groups, the bundles varying in the decreasing direction, a diagram was created in which the corresponding stresses were connected (FIG. 10).

Incorporating the plots of the nine overlapping co-expressed gene bundle maps above, this plot shows the initial relationship specific to the interrelated gene bundles and stresses commonly used between individual stresses, as well as the association between stresses and individual gene bundles. It has the advantage that the reaction gene bundle can be identified in more detail. Therefore, it is arbitrarily called a cross-talk map.

At this time, a bundle of genes having four different origins can be integrated into one to more accurately identify the correlation.

< Example 5> Selected Correlation ( cross-talk ) How to determine the authenticity of a gene bundle

Enrichment of a certain level or more as a result of using the calculation method devised by Subramanian A et al (PNAS October 25, 2005, vol. 102, No. 43, 15545-15550) to select a set of genes that respond under specific conditions. Genes that play a decisive role (hereinafter, referred to as key genes) are determined as a bundle of genes having a different expression pattern from the normal state according to the stress. Here, it is easy to see that even though a set of genes activated under all five stress conditions are not identical, the key genes that are thought to be actually associated with the stress are not identical. Thus, as the association stresses and connectivity of gene bundles in a cross-talk map alone cannot determine their correlated genetics, it is necessary to determine a true cross-talk gene bundle. More precise selection methods are needed.

If expression is enhanced in five or more stress conditions and judged to be a gene bundle candidate judged to be a correlated gene, a verification process is needed to determine the authenticity at the gene level. Focusing on the core genes of the genes included in the bundle for this verification, the contribution of gene bundles that participate in the stress response to the ratio of the core genes to each of the stresses is the overlapping core among the stresses. When the ratio of genes is defined as the cohesion between genes belonging to the gene bundle (Fig. 12), and represented by the large and the small of these values, each of the correlation candidate gene bundles obtained as a result of the selection is one of the following four cases. Would be one. The first is a group of genes with high cohesion and high contribution that correlate to all stimuli; the second is a group of genes with very high contribution and cohesion for some of the stimuli judged to have low overall cohesive but interrelationships; An association bundle, a cohesive gene with high cohesion but low contribution, can be divided into a fourth, a false positive gene bundle with low cohesion and low contribution.

If such a method is applied to a gene bundle determined as a faithful gene bundle candidate in all five or more stress conditions, the arrangement is as shown in FIG. 11A. Of these, DR55x57y was categorized in the first category, with a high 77% contribution to the associated stimulus and 70% of the total number of genes in that gene bundle commonly associated with four or more environmental stress responses (Figure 11B). Based on this result, a real cross-talk gene bundle was determined.

< Example 6> Determination of association of genes in gene bundle

Once the co-expression gene bundle map and faithful gene integration map have been used to identify trends in gene expression changes in a gene bundle unit, it is important to understand how the genes in that gene bundle actually correlate with associated stress. Interpretation at the genetic level is essential to understand. Here, in order to examine the practical relationship between the genes included in the co-expressed gene bundle, the genes belonging to the bundle were examined in DR55x57y, which was determined to be a gene bundle in which genes having a high need for stress response were collected in Example 5 Their expression profile was analyzed by applying a reverse engineering algorithm. We applied a reverse engineering algorithm called 'ARACNe' (Nature Genetics 37: 382-390), which reconstructs the cellular network from the microarray data developed by Califano lab at Columbia University, USA. The sub-regulatory network was configured, and the core gene for each stress was displayed therein, thereby obtaining the result of FIG. 14. As shown in FIG. 14, it was found that a large number of genes constituting the sub-regulartory network correspond to the core genes in each stress response in vivo.

As a result of the above results, it was found that the genes obtained through the method of discovering the related genes actually function as expected, and thus, the data analysis method of the present invention can be effectively used.

Figure 1 is a formula for calculating the relative positions of genes based on gene expression similarity is a graphical representation of the basic concept of the VxOrd algorithm.

FIG. 2 is a diagram showing 1436 bundles of data stored in a microarray database provided by TAIR using VxInsight, a data mining software, for constructing a bundle of genes co-expressed from mega clustering results of 28,000 Arabidopsis genes.

3 is a graph showing a method of preparing four kinds of gene bundle catalogs having different origins. It is classified into UL, UR, DL and DR according to the position of origin.

FIG. 4 is time series data designed to view gene expression changes over time by treating nine abiotic stresses on the roots and stems of Arabidopsis spp. And performing a microarray experiment on the individual after a certain time. .

5 is a diagram showing the change in the initial gene expression of the Arabidopsis to stress when the number of gene bundles showing a contrast expression difference at the time of each stress divided into roots and stems, showing a difference not only in the type of stress but also in the tissues. to be.

FIG. 6 shows overlapping co-expression maps that combine data analyzed by VxInsight with x / y positional coordinate values indicated by gene bundle names and the starting point of type code. This is a diagram visualizing changes in the expression of gene bundles by marking them at positions corresponding to enriched gene sets.

7 shows that blue is expressed in A-group and red is expressed in B-group. Different gene bundles are expressed differently according to stress (enrich) and the degree of similarity of responses is shown through the overall tendency of the gene bundle distribution indicated by co-expressed gene bundles.

FIG. 8 is a more detailed comparison of gene sets whose expression is enriched in each stress environment selected through GSEA. Horizontal refers to the type of stress, vertical refers to the name of the loyalty gene bundle.

Figure 9 shows gene bundles activated in the A-group when the root tissues are cold, genotoxic, heat, oxidative, salt, or wound stress. One of the genes that is activated in the B-group when one is 'DL_38x19y' and drought, genotoxic, osmotic, salt or UVB stress Signal intensity on gene expression profiles using a visualizing program called 'tree-view' for the time-domain expression profile corresponding to the initial (30min, 1h, 3h) of the genes belonging to 'DL_55x55y' ) Is a diagram showing the degree of image.

10 is a bundle that changes the whole loyalty gene bundle only in the direction of increasing expression, the bundle only changes in the direction of decreasing, using the graph visualization program 'Graphviz' bundles that change in the direction of increasing or decreasing with stress After dividing into 3 groups, these stresses are connected.

11 (a) shows that the core genes, which are the genes that played a decisive role in selecting the co-expression gene set and the co-expression gene set selected as faithful gene bundles by receiving a fidelity of a predetermined score or more, are activated. This is a table of stress by letting. (b) Table showing GSEA analysis values for DR55x57y, one of the gene bundles.

12 is a contribution of a bundle of genes participating in a stress response to a ratio of core genes to stress. Figures are defined and illustrated in terms of cohesion.
12A refers to a group of genes having a cohesion of 0.5 or more, but a contribution of less than 0.5, which are correlated with all stimuli,
B of FIG. 12 refers to a group of genes having a truly mutually related cohesion and contribution of 0.5 or more,
C in Figure 12 refers to a group of genes that are false positive cohesion and contribution less than 0.5,
D in FIG. 12 refers to a group of genes whose cohesion with similar correlations is less than 0.5, but the contribution is greater than 0.5.

FIG. 13 is a diagram illustrating that gene bundles that are strongly expressed in different stress conditions are divided into four types in terms of contribution and cohesion.

FIG. 14 illustrates a sub-regulatory network of genes belonging to a strongly expressed gene set by applying a reverse engineering algorithm called 'ARACNe', and a core gene for each stress. (core gene) is shown.

15 is a schematic diagram showing the flow of the present invention.

Claims (13)

A first step of inputting a gene obtained on the basis of gene expression similarity in the microarray data and a relative position in which the gene is expressed in two-dimensional coordinates by using a data mining program for determining the relative position of genes; A second step of determining a boundary size for classifying the two-dimensional coordinates of the gene input in the first step into a gene bundle; A third step of creating a new gene bundle catalog having four different origins based on the boundary size of the gene bundle determined in the second step; Statistical analysis is applied to the gene bundle catalog created in the third step, microarray data is obtained for normal cells (control) and stimulated cells (experimental group), and then the microarray data is compared and analyzed for genes. A fourth step of displaying an expression aspect; Based on the analyzed data of the fourth step, using a visualization program loaded into the microprocessor and the memory of the computer, the data associated with all the analyzed gene bundles of the fourth step are visualized to select the correlated gene bundles. Method of discovering a large amount of genes involved in regulating the function of the cells in the microarray data comprising a fifth step. The microarray data of claim 1, wherein the microarray data is selected from the group consisting of a TAIR database, a gene expression omnibus (GEO) of NCBI, and an array express of EBI. How to excavate. The method of claim 1, wherein the data mining program is VxOrd or STN AnaVist. The method of claim 1, wherein the size of the boundary of the second step is 10 to 20. 3. 5. The method of claim 4, wherein the size of the boundary is 15. 6. The method of claim 1, wherein the origin is located at a position consisting of 1/2 position of the X axis, 1/2 position of the Y axis and 1/2 position of the X and Y axes based on the boundary size of each gene bundle. A method for discovering large amounts of genes involved in regulating cell function in microarray data. The method of claim 1, wherein the statistical analysis method is gene set enrichment analysis (GSEA) or parametric analysis of gene set enrichment (PAGE). . The method of claim 1, wherein the graph visualization program is Graphviz or pajek. A first step of inputting a gene obtained on the basis of gene expression similarity in the microarray data and a relative position in which the gene is expressed in two-dimensional coordinates by using a data mining program for determining the relative position of genes; A second step of determining a boundary size for classifying the two-dimensional coordinates of the gene input in the first step into a gene bundle; A third step of creating a new gene bundle catalog having four different origins based on the boundary size of the gene bundle determined in the second step; Statistical analysis is applied to the gene bundle catalog created in the third step, microarray data is obtained for normal cells (control) and stimulated cells (experimental group), and then the microarray data is compared and analyzed for genes. A fourth step of displaying an expression aspect; Based on the analyzed data of the fourth step, using a visualization program loaded into the microprocessor and the memory of the computer, the data associated with all the analyzed gene bundles of the fourth step are visualized to select the correlated gene bundles. The fifth step; For the gene bundle selected in step 5, the distribution of core genes in the bundle according to the stimulus is classified into cohesion and contribution, and these gene bundles are correlated to all stimuli according to the degree of response to the stimulus. Coherent gene bundles with cohesion and contribution of 0.5 or more ii) bundles of genes with cohesiveness of less than 0.5 but contributions of 0.5 or more A sixth step of finding real cross-talk gene bundles after four subdivisions, such as false positive gene bundles all less than 0.5; And Reconstruct intracellular networks using a program that implements a reverse algorithm loaded in a microprocessor and memory for genes included in the real cross-talk gene bundle found in step 6 above. A method for screening interrelated genes comprising a seventh step of selecting a gene that is predicted to be relevant. 10. The method of claim 9, wherein the program implementing the reverse algorithm is ARACNe. An input unit for receiving relative position data of a gene based on expression similarity of a gene in microarray data and a gene represented by two-dimensional coordinates of the gene; A first calculating unit classifying the two-dimensional coordinates of the input gene into gene bundles according to a predetermined boundary size; A second operation unit classifying a new gene bundle catalog having four different origins based on the boundary size of the gene bundle; A third calculator configured to statistically analyze the gene bundle catalog to obtain microarray data of normal cells (control) and stimulated cells (experimental group), and then compare the values to display expression patterns of genes; A fourth calculating unit which visualizes the bundle of genes obtained by the first calculating unit in an expression map based on data obtained by the third calculating unit, and selects the correlated gene bundles by using a visualization program loaded in a microprocessor and a memory of the computer; And Mass excavation apparatus for interrelated genes involved in regulating the function of the cell configured as an output means for outputting the result of the cross-associated gene bundle screening. An input unit for receiving relative position data of a gene based on expression similarity of a gene in microarray data and a gene represented by two-dimensional coordinates of the gene; A first calculating unit classifying the two-dimensional coordinates of the input gene into gene bundles according to a predetermined boundary size; A second operation unit classifying a new gene bundle catalog having four different origins based on the boundary size of the gene bundle; A third calculator configured to statistically analyze the gene bundle catalog to obtain microarray data of normal cells (control) and stimulated cells (experimental group), and then compare the values to display expression patterns of genes; A fourth operation unit which visualizes the gene bundles obtained by the first operation unit using an expression map based on the data obtained by the third operation unit, using a visualization program loaded in a microprocessor and a memory, and selects the correlated gene bundles; Based on cohesion and contribution to gene bundles that respond to multistimulation Iii) a bundle of truly correlated genes with at least 0.5 cohesion and a contribution that correlates with all stimuli ii) a bundle of genes with a cohesion of less than 0.5 but a contribution of at least 0.5 Bundle iii) a fifth operation unit classifying false positive gene bundles having both cohesion and contribution ratios less than 0.5 and finding true correlated gene bundles; And And a sixth operation unit configured to reconfigure the intracellular network for genes included in the true correlation gene bundle found by the fifth operation unit and to select genes that are predicted to be correlated in the cell. Screening device for related genes. 13. The reciprocal of claim 12, wherein the multistimulus is any one or more abiotic stresses selected from cold, salt, ultraviolet stress, drought, osmotic pressure, wound, genotoxicity, oxidation and heat. Screening device for related genes.

Images