JP2006323830A - System for searching candidate gene related to phenotype - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To identify, from a group of genes unlinked with various types of information on diseases and so on, a new gene or a candidate gene having a relation to any of such information. <P>SOLUTION: This system comprises a score computing means and an evaluation value computing means. Using a training set comprising groups of genes that are already known to be related to information linked with genes and including gene-identifying data for identifying a pertinent gene and a database associating the gene-identifying data with annotation data on the gene identified by the gene-identifying data, the score computing means computes a plurality of scores indicating a relation of the annotation data on the genes included in the database with the information. From the plurality of scores computed by the score computing means, the evaluation value computing means computes an evaluation value indicating a relation of the gene with the information. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a gene analysis apparatus capable of identifying a new gene related to specific information (for example, a disease) or a candidate gene related to the information using gene annotation information.

  Information on protein functions in the database has been described in each protein database using a natural language, so that there is a problem that expression methods for the same function do not necessarily match between the databases. In order to solve this problem, recently, a database called Gene Ontology (hereinafter referred to as GO) has been constructed (http://www.geneontology.org/) that uniformly describes information on protein functions consistently. In GO, each function of proteins is expressed uniformly using a hierarchical structure.

  The G2D method (Perez-Iratxeta C et al. Nat Genet 2002 (31) 316-319: Non-Patent Document 1) is known as a method for searching for the relationship between gene functions and specific diseases using GO. It has been. The G2D method is based on “specific disease symptoms [information about C in MeSH term (http://www.nlm.nih.gov/mesh/meshhome.html)]” and “information about GO annotated to the causative gene”. The relationship between and is quantified by mediating the relationship between each element and the compound (MeSH D). At this time, the strength of the relationship to the disease is defined by the frequency of occurrence of the “specific relationship” with respect to “any relationship appearing in the paper”.

  In addition, the Phenotype Cluster method (Freudenberg J et al. Bioinformatics 2002 (18) S110-S115: Non-Patent Document 2) is known as a method for searching for the relationship between gene functions and specific diseases using GO. It has been. The Phenotype Cluster method mediates the relationship between "specific symptoms" and "information called GO annotated to the causative gene" by "relationship between each element and symptom characteristics (classification such as onset time and site)" By quantifying. At this time, the strength of the relationship is defined by the appearance frequency of the “specific relationship” with respect to the “relationship that appears in all the disease-related genes”.

  Furthermore, the POCUS method (Turner FS et al. Genome Biol 2003 (4) R75: Non-Patent Document 3) is known as a method for searching for the relationship between gene functions and specific diseases using GO. Yes. The POCUS method uses the relationship between `` specific symptoms '' and `` information called InterPro Identifier (http://www.ebi.ac.uk/interpro/) annotated to the causative gene and GO '' as a result of simulation. The strength of the relationship is defined by the frequency of occurrence of the “specific relationship” with respect to the “occurrence by chance”.

  However, these methods have two major problems as follows. First, it relies heavily on limited information called GO. In other words, many annotations are added to genes other than GO, but information other than GO annotated to genes is not used. Second, the information to be compared to define the strength of a specific relationship is very limited information. The G2D method uses “relationships that appear in papers”, the Phenotype Cluster method uses “relationships that appear in disease-related genes”, and the POCUS method uses simulations as comparison targets, but “all or almost all genes” as comparison targets. Is not done.

  In addition, these methods cannot identify a novel gene associated with a specific disease or a candidate gene associated with a specific disease from a gene group that has not been shown to be associated with a specific disease.

Perez-Iratxeta C et al. Nat Genet 2002 (31) 316-319 Freudenberg J et al. Bioinformatics 2002 (18) S110-S115 Turner FS et al. Genome Biol 2003 (4) R75

  Therefore, in view of the actual situation as described above, the present invention provides a gene analysis apparatus that can identify a novel gene or a candidate gene related to the information from a group of genes not associated with various information such as diseases. The purpose is to provide.

The present invention that has achieved the above-described object includes the following.
(1) A training set that includes gene identification data that is known in advance to be related to information related to genes, includes gene identification data that identifies the genes, and genes identified by the gene identification data and the gene identification data. A score calculation means for calculating a plurality of scores indicating the relevance with the information for the annotation data of the genes included in the database, and a plurality of scores calculated by the score calculation means. A gene information analyzing apparatus comprising: an evaluation value calculating means for calculating an evaluation value indicating a relationship between the gene and the information from the score.
(2) The score calculation means, for annotation data associated with a score calculation target gene, from the appearance frequency of the annotation data included in the training set and the appearance frequency of the annotation data of the gene included in the data set The apparatus according to (1), wherein a score is calculated.
(3) The apparatus according to (1) or (2), wherein the score calculation means calculates the score for annotation data of all genes included in the database.
(4) The score calculating means calculates the score based on the homology between the base sequence data of the gene to be score-calculated and the base sequence data of the genes constituting the training set (1) Equipment.
(5) The score calculation means calculates the correlation coefficient between the expression pattern data of the gene to be score-calculated and the expression pattern data of the genes constituting the training set from the parameters included in each expression pattern data. The apparatus according to (1), wherein a score is calculated based on the correlation coefficient.
(6) The gene identification data included in the training set is associated with a basic value quantifying the relationship between the gene specified by the gene identification data and the information described above (1) Equipment.
(7) Result output means for associating and outputting the gene identification data included in the data set excluding the training set from the data set included in the database and the evaluation value calculated for the gene specified by the gene identification data The apparatus according to (1), further comprising:
(8) The evaluation value calculation means calculates an evaluation value for determining whether the gene for which the score is to be calculated is close to the gene group constituting the training set or the gene group constituting the data set excluding the training set The device according to (1), characterized in that:
(9) The evaluation value calculating means calculates, for the score calculation target gene, the Mahalanobis general distance to the gene group constituting the training set and the Mahalanobis general distance to the gene group constituting the data set excluding the training set. The apparatus according to (1), characterized in that an evaluation value is calculated as a difference value of each Mahalanobis general distance.
(10) The device according to (1), wherein the information related to the gene is disease type information.
(11) The score calculation means is at least one selected from nucleotide sequence homology data, InterPro ID data, enzyme number data, metabolic pathway data, molecular function data in GO data, biological process data, subcellular component data, and expression pattern data The apparatus according to (1), wherein a score is calculated for the data of
(12) The score calculation means obtains the score for the annotation data associated with the gene included in the training set, stores the score in the storage means as a table associated with the annotation data, and stores the score in the score calculation target gene. The apparatus according to (1), wherein the table is searched by using the associated annotation data as a key, the score is read, and the read score is calculated as a score of the annotation data related to the score calculation target gene.

  The present invention can also be applied to a gene information providing apparatus that provides an analysis result data set constructed by the above gene analysis apparatus. That is, a gene information providing apparatus to which the present invention is applied is constructed by a management means for managing information related to genes, which is input via an input means or a communication network, as input data, and the gene information analyzing apparatus. A gene information providing apparatus comprising a result output means for outputting an analysis result data set.

  According to the gene analysis apparatus according to the present invention, it is possible to identify a new gene or a candidate gene having relevance to the information from a gene group not associated with various information such as a disease. In particular, according to the gene analysis apparatus of the present invention, annotation data associated with the gene group included in the training set, and annotation associated with the gene group included in the data set excluding the training set from all data sets. Since multiple scores are calculated from the data, comprehensive and objective analysis can be performed.

Hereinafter, the present invention will be described in detail with reference to the drawings.
As shown in FIG. 1, the gene analysis apparatus 1 according to the present invention is connected to a user terminal 3 via a communication network 2 such as the Internet, and is connected to the user terminal 3 via the communication network 2. Data exchange can be executed. Thereby, the gene analysis device 1 can provide the gene analysis result to the user terminal 3 via the communication line network 2. The user operates the user terminal 3 to input information related to the gene to the gene analysis device 1, and the gene analysis device 1 that has received the input information has a new relationship with the information based on this information. A gene group or a candidate gene group relevant to the information is provided to the user terminal 3.

  Here, information related to genes includes, for example, disease type information, ethnic and regional backgrounds, immune reaction information, etc. as information related to human-derived genes. Examples of information to be performed include pathologic information, various stress tolerance information, growth and breeding information, pest acquisition information, environmental adaptation information, and the like. In the case of plant-derived genes, especially for rice genes, phenotypes (multifactor phenotypes) caused by multifactors such as pollination, fertilization and fertility can be mentioned.

  In the following, human disease type information will be exemplified and described as information related to genes. That is, in the following description, the gene analysis device 1 provides a novel gene related to a human disease or a candidate gene related to the disease. However, the gene analysis device 1 according to the present invention is not limited to a device that provides a novel gene related to a disease or a candidate gene related to the disease, and can be applied to various information as described above. Can do.

  The disease information is not particularly limited, and examples thereof include multifactorial diseases such as rheumatoid arthritis, prostate cancer, pancreatic cancer, hypertension, type 2 diabetes, schizophrenia, and bronchial asthma. Moreover, it is preferable that the cause of the disease is mainly a disease such as protein mutation, chromosomal abnormality, or polymorphism.

  As shown in FIG. 2, the gene analyzing apparatus 1 includes a central processing unit 4 that performs operation control and data calculation of the entire apparatus, an HDD 5 that has a hard disk storing a program and various data to which the present invention is applied, information, Input means 6 such as a keyboard and mouse that can input program execution instructions, transmission / reception means 7 that transmits / receives information to / from an external terminal device or database via the communication network 2, and display means such as a display 8, a memory 9 in which various data and programs are recorded, and a storage means 10 storing various data.

  The gene analyzer 1 stores in the storage means 10 an analysis result data set including information related to genes (disease type information) and a new gene group related to the information or a candidate gene group related to the disease. Although it stores, you may have separately the database (not shown) which stores the said analysis result database. The analysis result data set may include an evaluation value that is associated with each gene included in the new gene group and / or the candidate gene group and indicates an association with a disease. Here, the gene analyzing apparatus 1 may be one in which the database is stored in the HDD 5 or the storage means 10, and if the database can be accessed via the transmission / reception means 7, the database is stored outside. You may have.

  The gene analyzer 1 accesses the database using the specific disease type information received by the transmission / reception means 7 as a key, reads the analysis result data set associated with the specific disease type information from the database, and outputs it to the user terminal 3 The program to be installed is installed.

  Hereinafter, a program for constructing an analysis result data set relating to specific disease type information (hereinafter referred to as this program) will be described. The gene analyzing apparatus 1 constructs a training set including a group of genes that are known in advance to be associated with the disease type information in the specific disease type information prior to the execution of this program. Storing. The training set can be constructed using, for example, another database in which disease type information and gene identification data are stored in association with each other. More specifically, the Online Mendelian Inheritance in Man (hereinafter referred to as OMIM) database storing human genes and hereditary diseases and the Entrez Gene database succeeding LocusLink can be used.

  When constructing a training set, first, by using these databases and searching for disease type information as a key, a known gene group related to the disease type information is specified. Next, using a database in which literature information is accumulated, the relevance to the disease is quantified for each gene constituting the gene group. Specifically, as shown below, a basic value indicating an association with a disease is calculated for each gene. The calculated basal values are associated with individual genes as a training set.

  The basic value of each gene is searched by searching a database storing literature information using gene identification data such as gene name and identification number as a key, and the relevance between the gene described in the hit document and the disease is evaluated, The result of the evaluation is quantified and calculated. The basic value calculated for each gene means a value obtained by objectively quantifying the relationship between a specific disease and the gene.

  More specifically, it is searched whether or not a search term phrase called “MeSH term (Medical Subject Headings term)” is included in the literature information associated with the searched gene using the disease type information as a key. For example, “Rheumatoid Arthritis” and “juvenile rheumatoid arthritis” are keyed as MeSH terms for rheumatoid arthritis, and “Prostatic Neoplasms” and “Intraepithelial Prostatic Neoplasia” are keyed as MeSH terms for prostate cancer.

  Further, since MeSH term is not attached to all abstracts, a search may be made using words other than MeSH term. For example, in the case of prostate cancer, the OMIM database may be searched using “prost *” and words related to cancer (for example, carcinoma, tumor, cancer, adenocarcinoma, neoplasm, etc.) as search words.

  As a result of the search, gene identification data (gene name or gene identification number) having relevance to the disease type information and literature information thereof can be obtained. Next, using the obtained literature information as a key, a database (so-called PubMed database) storing the literature information is searched, and a document describing the gene is extracted. If a plurality of gene identification data are obtained as a result of the search, documents are extracted for all genes.

  After that, in each extracted document, it is confirmed how much the target gene is described as having a relationship with the actual disease. A score evaluation value indicating the relationship between the target gene and the disease is set for each document extracted based on the result of the confirmation. For example, when extracting prostate cancer-related genes, 297 gene identification data (OMIM ID) can be obtained from the OMIM database using “Prostatic Neoplasms” and “Intraepithelial Prostatic Neoplasia” as keys. The OMIM database also includes literature information that does not show a direct relationship to a disease, such as gene cloning information from a disease cDNA library and gene information localized in a disease susceptibility region. Therefore, it can be evaluated that these literature information shows a false positive when evaluating as a causative gene of prostate cancer. In addition, for example, the KANGAI 1 (OMIM ID: 600623) gene has already been shown to have an effect on prostate cancer. However, Ichikawa T et al. Cancer Res 1991 (51 3788-3792, Ichikawa T et al. Cancer Res 1992 (52) 3486-3490) does not show the association. As the score evaluation value, for example, 1 to 3 may be set for documents that affirm relevance, 0 for documents that do not mention relevance, and negative for documents that clearly deny relevance. it can.

  Next, by calculating the average of the score evaluation values set for each document information, a basic value indicating the relationship between the disease and the specific gene is obtained. When n pieces of literature information are searched for a certain gene and score evaluation value Xi (i = 1 to n) is set for each piece of literature information, the basic value

は、下記（e1）式で表される（ａは疾患との関連性を有するものとして検索された遺伝子数を意味する。）。

Is represented by the following formula (e1) (a means the number of genes retrieved as having an association with a disease).

  The genetic analyzer 1 stores training sets including basic values for various diseases, for example, in a hard disk or a database. That is, the gene analyzing apparatus 1 has a training set for each of various information related to genes as described above. A functional configuration of the gene analyzer 1 is shown in FIG. That is, this program causes the apparatus having the hardware resources shown in FIG. 2 to function as the score calculation means 20, the evaluation value calculation means 21, and the analysis result output means 22. In addition, the gene analyzing apparatus 1 is connected to a human gene database 23 in which various annotation data are integrated. As shown in FIG. 4, the data included in the training set includes gene identification data for identifying individual genes in the human gene database 23 used in the program, and the above-described basics related to genes identified by the gene identification data. It consists of a value and

  On the other hand, the gene analyzing apparatus 1 has the above human gene database 23 or can access it to read data in the database. The human gene database 23 that can be used is not particularly limited, and includes H-Invitational Database (hereinafter, H-InvDB). H-InvDB is described in detail in Imanishi T et al. PLoS Biol 2004 (2) 856-875, and is a database available via the Internet. This H-InvDB contains GO information, gene structure, function, alternative splicing isoform, non-coding region functional RNA, functional domain, subcellular localization, SNP mapping and microsatellite for registered genes. Motifs, gene expression profiles, comparison results with other species, etc. are stored in an integrated manner as annotation data.

  This program causes a computer device having the training set and having access to the human gene database 23 to function as the score calculation means 20 first. The score calculation means 20 in the gene analyzer 1 uses the above-described training set and a training set using all the data sets (hereinafter referred to as a reference data set) included in the human gene database 23 (for example, H-InvDB). The score is calculated for each annotation data associated with the gene included in the.

  Note that the annotation data used for the calculation depends on the annotation items stored in the database to be used. In this example, H-InvDB is illustrated, but a database having different annotation items can also be used. When a database having different annotation items is used, the score calculation method described below is also different.

  Here, the score is a value indicating whether the annotation data associated with the gene included in the training set indicates the specificity of the training set constructed for each disease type information. Although the score to calculate is illustrated below, the technical scope of this invention is not limited to the following scores.

Score human gene database 23 based on the homology of the gene sequence include annotation data associated with the homology as an annotation data integrated for a given gene. Specifically, for a given gene, so-called e-values relating to homology with other genes are associated as annotation data. Examples of other genes include a Paralogous gene. A collateral gene is a gene in the same species that is caused by duplication and the resulting mutation from a common ancestor. Due to its high homology, it is thought to have high protein function similarity with the other gene caused by duplication.

  In calculating this score, a side gene for a gene included in the training set was defined as an e-value included in the human gene database 23 exceeding a threshold (for example, 100). That is, when the e-value is less than the threshold, the score is not calculated instead of the side gene.

Specifically, first, the human gene database 23 is searched using the gene identification data included in the training set as a key, and a side gene (e-value greater than or equal to a predetermined threshold) associated with the gene specified by the gene identification data. And the e-value associated with the neighboring gene are read out. Next, the read e-value and the basic value associated with the gene having the gene identified by the read gene identification data as a side gene are read, and the score (S _{p, x} ) is calculated according to the following equation.

The score calculation means 20 calculates this score (S _{p, x} ) for all the parasite genes associated with all the genes included in the training set. The obtained score (S _{p, x} ) is stored in a storage means or the like as a table associated with gene identification data indicating a side gene. When the genes included in the reference data set are stored as side genes in the table, the score (S _{p, x} ) is the same as the genes included in the training set. It shows how much similarity there is.

  Next, the score calculation means 20 searches the table using the gene identification data of the genes included in the reference data set as a key, and when the gene is included as a side gene, the score stored in the table is calculated. , Set as the score of the gene. In addition, the score calculating means 20 sets the score of the said gene to 0, when the searched gene is not contained as a sideline gene.

Score human gene database 23 based on the functional annotation information gene annotation data about the function of the gene (hereinafter, referred to as functional annotation information) it is associated. Functional annotation information includes InterPro ID, enzyme number (EC), and metabolic pathway (KEGG pathway). The InterPro ID is identification data included in a database provided by the European Bioinformatics Institute, and is assigned to each functional name of the protein. The enzyme number (EC) is a classification number defined by the International Biochemical Union, and is assigned a 4-digit number (for example, EC 2.3.2.13) according to function. The metabolic pathway (KEGG pathway) is identification data included in a database provided by the Kyoto Encyclopedia of Genes and Genomes, and is assigned to each molecular mutual network in a biological process such as a metabolic pathway.

  This score is a score calculated from the appearance frequency of the function annotation information in the training set and the appearance frequency of the function annotation information in the reference data set for the function annotation information associated with the genes included in the training set. . For example, when three types of function annotation information, ie, InterPro ID, enzyme number (EC), and metabolic pathway (KEGG pathway) are associated with each gene, the score calculation means 20 sets the score for each type of function annotation information. Is calculated.

  Specifically, first, the human gene database 23 is searched using the gene identification data included in the training set as a key, and the function annotation information associated with the gene specified by the gene identification data is read out. Next, the score calculation means 20 searches the human gene database 23 using the gene identification data included in the training set and the reference data set as a key when calculating the score for the predetermined function annotation information among the read function annotation information. Calculate the following numbers.

  Moreover, the score calculation means 20 calculates the average value of the basic values in the gene group associated with the predetermined function annotation information. The average of this base value

と表すと、本スコアは下記式(e3)で与えられる。

This score is given by the following formula (e3).

The score S _{f, n} calculated by the score calculation means 20 indicates the specificity of each function annotation information in a specific disease. It is clear that if the score S _{f, n} is high and the specificity is high, it has an important role in the disease. According to the above formula (e3), it is possible to calculate a value that faithfully evaluates the point of specificity without giving a high score to the function annotation information that exists frequently with low specificity. .

When a plurality of pieces of data are associated with a predetermined type of function annotation information, the score calculator 20 calculates a score S _{f, n} for each data. For example, when three InterPro IDs are associated with a predetermined gene included in the training set, S _{f, n} is calculated and summed for each InterPro ID.

The score calculation means 20 calculates this score S _{f, n} for all function annotation information associated with all genes included in the training set. The obtained score S _{f, n} is stored in the storage means or the like as a table associated with the function annotation information.

  Next, the score calculation means 20 searches the table using the function annotation information of the gene included in the reference data set as a key, and if the function annotation information is included, the score stored in the table is Set as the score for the gene. When a plurality of function annotation information (for example, three InterPro IDs) is associated as a predetermined type of function annotation information (for example, InterPro), the predetermined type of function annotation information is used as the score of the gene. The score for is calculated as the sum of each score (e4). The score calculation means 20 sets the score for the gene to 0 when the searched function annotation information is not included in the table.

In this way, the score calculation means 20 calculates the score S _{f, n} for each of a plurality of types of function annotation information such as InterPro ID, enzyme number (EC), and metabolic pathway (KEGG pathway) for the genes included in the reference data set. It can be calculated.

  By the way, the human gene database 23 is associated with GO (Gene Ontrogy) data as function annotation information. GO data is roughly divided into three categories: (1) molecular function, (2) biological process, and (3) subcellular component, and is associated with gene identification data. Here, the GO data has a hierarchical structure in each category, and has subdivided information as it is located in a lower layer.

  The score calculation means 20 defines a score in consideration of the hierarchical structure when assigning a score to GO data. For example, if you compare the definition of cell (GO: 0005623) in the top layer of subcellular components and the definition of nucleus (GO: 0005634) in the third layer, clearly “nucleus” located in the lower layer is more detailed Information.

  Therefore, the score calculation means 20 uses the following equation (e5) that takes into account the GO hierarchical structure based on the above equation (e3) as the score for GO data.

In the above formula (e5), Ph _{, n} is a position in the hierarchical structure, that is, a value that increases as the hierarchy is lower.

The score calculation means 20 calculates this score _{Sgo1-3, n} for all GO data associated with all genes included in the training set. The obtained score S _{go1-3, n} is stored in the storage means or the like as a table associated with the GO data.

Next, the score calculation means 20 searches the table using the GO data of the gene included in the reference data set as a key, and if the GO data is included, the score S _go1-3 stored in the table is stored. _{, n} is set as the score for the gene. When a plurality of function annotation information (for example, three molecular functions) is associated as a predetermined type of GO data (for example, molecular function), the predetermined type of GO data is used as the score of the gene. The scores for are calculated and summed up as the sum of each score (e6). The score calculation means 20 sets the score for the gene to 0 when the searched GO data is not included in the table. In this way, the score calculation means 20 can calculate the score S _{f, n} for each GO data for the genes included in the reference data set.

Score human gene database 23 on gene expression patterns, the annotation data related to gene expression patterns (hereinafter, referred to as expression pattern data) is associated.

  Several cluster analyzes have shown correlation between functional similarity between genes based on GO data and similarity in gene expression patterns (Eisen MB et al. Proc Natl Acad Sci 1998 (95) 14863- 14868, Lagreid A et al. Genome Res 2003 (5) 965-979). From these findings, it is considered that genes having similar functions are subjected to similar gene expression control. Finding similarities in gene expression patterns based on this hypothesis suggests that they are under the same gene expression control and may share similar functions between the genes.

The score calculation means 20 first reads the expression pattern data associated with the genes included in the training set and the expression pattern data associated with the genes included in the reference data set, and calculates the correlation coefficient between these expression pattern data. To do. Specifically, the score calculation means 20 uses the standard deviation δ _x of the expression pattern data X associated with the genes included in the training set and the standard deviation δ _y of the expression pattern data Y associated with the genes included in the reference data set. Using the covariance Cov (X, Y) of the expression pattern data X and the expression pattern data Y, the correlation coefficient Rxy can be calculated by the following equation. The expression pattern data are iAFLP data (Kawamoto S et al. Genome Res 1999) stored in H-ANGEL (Tanino M et al. Nucleic Acids Res 2005 (33) D567-D572), a sub-database of H-InvDB. (12) 1305-12, Sese J et al. Nucleic Acids Res 2001 (29) 156-8) 40 tissue classification can be used.

  In the above formula, n means the number of data, Xi and Yi mean the expression levels in the i-th tissue of the expression patterns X and Y, respectively, and μx and μy respectively in the i-th tissue in each expression pattern. Means the average expression level.

Next, the score calculation means 20 stores a table that associates the gene identification data associated with the expression pattern data X, the gene identification data associated with the expression pattern data Y, and the obtained correlation coefficient Rxy. And so on. Next, the score calculation means 20 searches the maximum correlation coefficient Rxy associated with the gene identification data with reference to the table using the gene identification data of the genes included in the reference data set as a key. The score calculation means 20 reads the basic value of the gene associated with the gene identification data with the gene identification data included in the training set associated with the maximum correlation coefficient Rxy as a key. Then, the score calculation means 20 calculates the score Se _{, x} according to the following formula (e7) for the genes included in the reference data set, using the read basic value and the correlation coefficient Rxy indicating the maximum.

In the above description, the score _{Se, x} is calculated using the expression pattern data associated with the genes included in the training set and the reference data set as the score for the gene expression pattern. Even if is associated with a cluster composed of a plurality of genes, the score _{Se, x} can be calculated similarly. A cluster means a group classified based on gene sequence information. In this case, the score calculation means 20 stores a table in which the cluster data associated with the expression pattern data X, the cluster data associated with the expression pattern data Y, and the obtained correlation coefficient Rxy are stored in the storage means or the like. Store. Here, cluster data means data for identifying a cluster. Then, the score calculation means 20 searches the maximum correlation coefficient Rxy associated with the cluster data with reference to the table using the cluster data to which the gene included in the reference data set belongs as a key. Thereafter, the score _{Se, x} can be calculated in the same manner.

As described above, the score calculation means 20 can calculate a plurality of scores for genes included in the reference data set using the gene identification data and annotation data included in the training set and the reference data set. More specifically, the score calculation means 20 can calculate the following eight types of scores as a plurality of scores in the above-described example.
1. 1. Score based on gene sequence homology Score based on gene function annotation information (InterPro ID)
3. Score based on gene function annotation information (enzyme number (EC))
4). Score based on gene function annotation information (KEGG pathway)
5. Score based on gene function annotation information (GO data (molecular function))
6). Score based on gene function annotation information (GO data (biological process))
7). Score based on gene function annotation information (GO data (subcellular component))
8). Score based on gene expression pattern

  In the above example, in calculating each score, the score calculation means 20 constructs a table including scores calculated in advance for annotation data associated with genes included in the training set, and uses the table to calculate the score. Each score was calculated for the gene of interest. However, in the present invention, the score calculation means 20 can also individually calculate annotation data associated with the genes included in the reference data set without constructing such a table in advance.

  For example, when calculating the scores 2 to 7, the score calculation means 20 first searches the human gene database 23 using the gene identification data included in the reference data set as a key, and the gene specified by the gene identification data. Read the function annotation information associated with. Next, the score calculation means 20 searches the human gene database 23 using the gene identification data included in the training set and the reference data set as a key,

を演算する。そして、スコア演算手段２０は、上記の(e3)式に従ってスコアを演算することができる。この場合、スコア演算手段２０は、スコア演算対象の遺伝子について直接演算することとなる。

Is calculated. And the score calculation means 20 can calculate a score according to said (e3) Formula. In this case, the score calculation means 20 directly calculates the score calculation target gene.

Evaluation Value Calculated from Score Next, in this program, the evaluation value calculating means 21 uses the plurality of scores calculated by the score calculating means 20 to determine the relationship between the genes included in the reference data set and the disease type information. The evaluation value shown is calculated. Here, the evaluation value is a value that is calculated for the genes included in the reference data set and indicates the distance to the gene group included in the training set. More specifically, the evaluation value calculation means 21 calculates an evaluation value using a discriminant function that can handle the above eight types of scores as having different attributes. Here, the discriminant function means a function that determines to which group each data belongs based on several variables.

  Specifically, the evaluation value is the distance from the gene included in group 2 to group 1 (MD1), where the gene group included in the training set is group 1 and the data set obtained by removing the training set from the reference data set is group 2. And the distance (MD2) to group 2 is calculated and calculated by the formula (MD2 / MD1). MD1 and MD2 can be calculated using Mahalanobis's general distance (Mardia KV et al. Multivariate analysis 1979 Academic Press, New York) that can take dispersion into account. The general distance to each group can be calculated according to the following formula (e8).

上記式(e8)において、Ｘはスコア演算手段２０で演算された複数のスコアの値であり、

In the above formula (e8), X is a plurality of score values calculated by the score calculating means 20,

は、トレーニングセット或いは参照データセット（例えばk個の遺伝子からなる）の平均である。なお、j = 1，2， ... ，kである。また、

Is the average of a training set or a reference data set (eg, consisting of k genes). Note that j = 1, 2,..., K. Also,

は、各群の分散共分散行列の逆行列である。

Is the inverse of the variance-covariance matrix of each group.

  Finally, the evaluation value calculation unit 21 stores the evaluation value obtained as described above in the storage unit as an analysis result data set relating to specific disease type information in association with the gene identification data. As described above, according to this program, an analysis result data set relating to specific disease type information can be constructed. The gene analyzing apparatus 1 may include an analysis result output unit 22 that outputs an analysis result data set constructed by this program.

  The gene analyzing apparatus 1 can similarly construct analysis result data sets for information related to various genes, and store the obtained analysis result data sets in the storage means in association with the information. Moreover, the gene analysis apparatus 1 includes the analysis result output unit 22 and is connected to a network as shown in FIG. 1, thereby providing a desired analysis result data set to the user terminal 3. . In the analysis result data set provided to the user terminal 3, a group of genes whose relevance to a specific disease is unknown is displayed as a list together with an evaluation value indicating the relevance to the disease. Based on the displayed list, the user can identify new genes that are related to the disease, or can identify candidate genes that are highly likely to be related to the disease. As described above, the gene analyzing apparatus 1 can also be realized as a gene information providing apparatus capable of providing useful information such as a novel gene or a candidate gene having a relation to a disease to a user.

Actual analysis results Using the above-described invention, a search was made for gene candidates (new disease susceptibility genes) involved in rheumatoid arthritis, which is a multifactorial disease and has severe symptoms, and prostate cancer. Specifically, a training set for rheumatoid arthritis was first constructed using the OMIM database as of August 2004. Table 1 shows a training set obtained by associating the gene identification data (gene name and gene symbol) with the basic value.

  Subsequently, this program was applied, and the above-mentioned eight types of scores were calculated for all data sets included in H-InvDB (human gene database 23). Then, applying this program, the group of genes included in the training set is group 1, the data set obtained by removing the training set from H-InvDB is group 2, and the distance of the genes included in group 2 to group 1 (MD1 ) And the distance to group 2 (MD2), and the evaluation value was calculated by the formula (MD2 / MD1).

  As a result, 469 genes were extracted as novel disease susceptibility genes for rheumatoid arthritis. Table 2 shows gene identification data (gene name, gene symbol and accession number), evaluation values (MD2 / MD1), MD1 and MD2 for the top 100 genes among the extracted 469 genes.

  That is, by applying this program, a new disease susceptibility gene group of rheumatoid arthritis shown in Table 2 was specified. When Table 2 is examined in detail, it can be noted that IL1RN, which is an interleukin 1 receptor antagonist, has been identified. IL1RN was not associated with rheumatoid arthritis in the OMIM database as of August 2004 when the training set was established. Subsequently, in December 2004, Horai et al. Reported an association between IL1RN and rheumatoid arthritis (Horai R et al. J Clin Invest 2004 114 (11) 1603-11). In other words, the effectiveness of this program could be demonstrated, such that this program can actually identify new genes related to specific disease type information.

  Similarly, a training set for prostate cancer was constructed using the OMIM database as of August 2004. Table 3 shows a training set obtained by associating the gene identification data (gene name and gene symbol) with the basic value.

  Subsequently, this program was applied in the same manner as in the above example to determine MD1, MD2, and evaluation values (MD2 / MD1). As a result, 357 genes were extracted as novel disease susceptibility genes for prostate cancer. Table 4 shows gene identification data (gene name, gene symbol and accession number), evaluation value (MD2 / MD1), MD1 and MD2 for the top 100 genes among the extracted 357 genes.

  Examining Table 4 in detail, it can be noted that EPHB2, a tyrosine kinase receptor, has been identified. EPHB2 was not associated with prostate cancer in the OMIM database as of August 2004 when the training set was established. Subsequently, an association between EPHB2 and prostate cancer was reported by Huusko et al. In September 2004 (Huusko P et al. Nat Genet 2004 (36) 979-983). In other words, the effectiveness of this program could be demonstrated, such that this program can actually identify new genes related to specific disease type information.

Although other examples to which the present invention is applied have been described by exemplifying human disease type information as information related to genes, the scope of the present invention is not limited to this example. This program can be applied to information related to genes derived from plants as information related to genes. As a specific example, a program for constructing an analysis result data set related to a rice multifactor phenotype will be described below.

  In this example, a training set consisting of a group of genes known in advance to be related to specific multifactor phenotype information in rice is constructed prior to execution of this program, and stored in, for example, the storage means 10 or the like. Yes. The training set in this example can be constructed using a database storing rice function annotation information. More specifically, Rice Annotation Project Data Base (RAP-DB), which is a rice gene annotation database, can be used.

  When constructing a training set, first, using these databases, a known group of genes related to the multifactor phenotype information is specified by searching for rice multifactor phenotypes as keys. Specifically, genes searched as follows are used as a training set. For example, when searching for genes related to pollination and fertilization, RAP-DB is searched using “Fertility” and “Sertility” as keys. In this example, unlike the case where the above-described human disease information is used, all the basic values given by the equation (e1) are 1. When rice genes are to be analyzed, even if a plurality of documents are retrieved as document information, there is no difference that gives a score evaluation value for each document. Therefore, in this example, it is not necessary to search for literature information about genes searched using the rice multifactor phenotype as a key. Alternatively, even if literature information on the gene is searched, it is not necessary to set a score evaluation value and calculate a basic value using the above equation (e1).

  In this example, the gene analyzing apparatus 1 stores training sets for various multi-factor phenotypes related to rice, for example, in a hard disk or a database. That is, the gene analyzing apparatus 1 has a training set for each of various information related to genes as described above. On the other hand, the gene analysis apparatus 1 has the rice gene database 23 or can access it to read data in the database. The human gene database 23 that can be used is not particularly limited, and includes RAP-DB. RAP-DB is described in detail in Ohyanagi H et al. NAR 2006 (34) D741-744, and is a database that can be used via the Internet. In the above example, H-InvDB is used. On the other hand, RAP-DB is used in this example, but there are annotation items that are common between these databases. Some annotation items are not included in the other database. The score calculation means 20 described below calculates various scores using annotation data depending on the annotation items stored in the database. In this example, the score is a value indicating whether the annotation data associated with the genes included in the training set indicates the specificity of the training set constructed for each multifactor phenotype. Although the score to calculate is illustrated below, the technical scope of this invention is not limited to the following scores.

Score rice gene database 23 based on the functional annotation information gene annotation data about the function of the gene (hereinafter, referred to as functional annotation information) is associated. Examples of function annotation information include InterPro ID, enzyme number (EC), and GO data as in the case of human genes. Specifically, first, the rice gene database 23 is searched using the gene identification data included in the training set as a key, and the function annotation information associated with the gene specified by the gene identification data is read. Next, the score calculation means 20 searches the rice gene database 23 using the gene identification data included in the training set and the reference data set as a key when calculating the score for the predetermined function annotation information among the read function annotation information. The calculation is performed using the above equations (e3, e4, e5).

The score calculation means 20 can calculate a plurality of scores for genes included in the reference data set using gene identification data and annotation data included in the training set and the reference data set. More specifically, the score calculation means 20 can calculate the following five types of scores as a plurality of scores in the above-described example.
1. Score based on gene function annotation information (InterPro ID)
2. Score based on gene function annotation information (enzyme number (EC))
3. Score based on gene function annotation information (GO data (molecular function))
4). Score based on gene function annotation information (GO data (biological process))
5. Score based on gene function annotation information (GO data (subcellular component))

Evaluation Value Calculated from Score Next, in this program, similar to the above-described example, the evaluation value calculating unit 21 uses a plurality of scores calculated by the score calculating unit 20 and the gene and many genes included in the reference data set. An evaluation value indicating the relevance with the information by factor phenotype is calculated. In this example, the evaluation value calculation means 21 calculates an evaluation value using a discriminant function that can handle the above five types of scores as having different attributes.

  In this example, the evaluation values are the same as in the above-described example, the group of genes included in the training set is group 1, the data set obtained by removing the training set from the reference data set is group 2, and the genes included in group 2 The distance to the group 1 (MD1) and the distance to the group 2 (MD2) are obtained and calculated by the formula (MD2 / MD1). MD1 and MD2 are calculated using Mahalanobis generalized distances that can take dispersion into account, and the generalized distances to the respective groups are calculated according to the above formula (e8).

  Also in this example, the evaluation value calculation means 21 stores the evaluation value obtained as described above in the storage means as an analysis result data set relating to specific multifactor phenotype information in association with the gene identification data. To do. As described above, according to the present program, it is possible to construct an analysis result data set relating to various types of multi-factor phenotype information related to rice. In this example as well, the gene analysis apparatus 1 may have a function of outputting an analysis result data set constructed by this program by the analysis result output means 22.

It is a schematic block diagram of the gene-analysis result provision system by the gene-analysis apparatus to which this invention is applied. It is a block diagram which shows typically the hardware constitutions of the gene-analysis apparatus to which this invention is applied. It is a block diagram which shows the functional structure of the gene-analysis apparatus to which this invention is applied. It is a block diagram which shows typically the data structure of the training set which the gene-analysis apparatus to which this invention is applied uses.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 ... Gene analysis apparatus, 2 ... Communication network, 3 ... User terminal, 4 ... Central processing unit, 5 ... HDD, 6 ... Input means, 7 ... Transmission / reception means, 8 ... Display means, 9 ... Memory, 10 ... Storage means , 20 ... score calculation means, 21 ... evaluation value calculation means, 22 ... analysis result output means, 23 ... human gene database

Claims (24)

A training set that includes gene identification data that is previously known to be related to information related to genes, includes gene identification data that identifies the gene, and gene annotation data and gene annotation data identified by the gene identification data And a score calculation means for calculating a plurality of scores indicating the relevance with the above information on the annotation data of the genes included in the database,
From a plurality of scores calculated by the score calculation means, an evaluation value calculation means for calculating an evaluation value indicating the relationship between the gene and the information,
A genetic information analysis apparatus comprising:   The score calculation means calculates a score for the annotation data associated with the score calculation target gene from the appearance frequency of the annotation data included in the training set and the appearance frequency of the annotation data of the gene included in the data set. The apparatus according to claim 1.   The apparatus according to claim 1 or 2, wherein the score calculation means calculates the score for annotation data of all genes included in the database.   2. The apparatus according to claim 1, wherein the score calculation means calculates a score based on homology between the base sequence data of a gene to be score-calculated and the base sequence data of genes constituting the training set.   The score calculation means calculates the correlation coefficient between the expression pattern data of the gene to be score-calculated and the expression pattern data of the genes constituting the training set from the parameters included in each expression pattern data, and the highest correlation coefficient The apparatus according to claim 1, wherein a score is calculated based on   The apparatus according to claim 1, wherein the gene identification data included in the training set is associated with a base value obtained by quantifying the relationship between the gene specified by the gene identification data and the information.   The apparatus further comprises a result output means for associating and outputting the gene identification data included in the data set excluding the training set from the data set included in the database and the evaluation value calculated for the gene specified by the gene identification data. The apparatus according to claim 1.   The evaluation value calculating means calculates an evaluation value for determining whether a gene for which a score is to be calculated is close to a gene group constituting the training set or a gene group constituting a data set excluding the training set. The apparatus according to claim 1.   The evaluation value calculating means obtains the Mahalanobis general distance to the gene group constituting the data set excluding the training set and the Mahalanobis general distance to the gene group constituting the training set for the score calculation target gene, The apparatus according to claim 1, wherein an evaluation value is calculated as a difference value of the Mahalanobis general distance.   The apparatus according to claim 1, wherein the information related to the gene is disease type information.   The score calculation means is for one or more data selected from base sequence homology data, InterPro ID data, enzyme number data, metabolic pathway data, molecular function data in GO data, biological process data, subcellular component data, and expression pattern data. The apparatus according to claim 1, wherein a score is calculated.   The score calculation means obtains the score for annotation data associated with a gene included in the training set, stores the score in a storage means as a table associated with the annotation data, and is associated with a score calculation target gene. 2. The apparatus according to claim 1, wherein the table is searched by using the annotation data as a key, the score is read, and the read score is calculated as the score of the annotation data related to the score calculation target gene. Management means for managing, as input data, information related to genes input via the input means or communication line network;
It consists of a group of genes known in advance to be related to the input information, and includes a training set including gene identification data for specifying the gene, gene identification data, and annotation data for the gene specified by the gene identification data. A score calculation means for calculating a plurality of scores indicating the relevance with the information for annotation data of genes included in the data set included in the database, and a plurality of scores calculated by the score calculation means. From the score, a result output means for outputting an analysis result data set constructed by a gene information analysis device comprising an evaluation value calculation means for calculating an evaluation value indicating the relationship between the gene and the information,
A genetic information providing apparatus comprising:   The score calculation means calculates a score for the annotation data associated with the score calculation target gene from the appearance frequency of the annotation data included in the training set and the appearance frequency of the annotation data of the gene included in the data set. The apparatus of claim 13.   The apparatus according to claim 13 or 14, wherein the score calculation means calculates the score for annotation data of all genes included in the database.   14. The apparatus according to claim 13, wherein the score calculation means calculates a score based on the homology between the base sequence data of the gene to be score-calculated and the base sequence data of the genes constituting the training set.   The score calculation means calculates the correlation coefficient between the expression pattern data of the gene to be score-calculated and the expression pattern data of the genes constituting the training set from the parameters included in each expression pattern data, and the highest correlation coefficient The apparatus according to claim 13, wherein a score is calculated based on the above.   14. The apparatus according to claim 13, wherein the gene identification data included in the training set is associated with a basic value obtained by quantifying the relationship between the gene specified by the gene identification data and the information.   The evaluation value calculating means calculates an evaluation value for determining whether a gene for which a score is to be calculated is close to a gene group constituting the training set or a gene group constituting a data set excluding the training set. The apparatus according to claim 13.   The evaluation value calculating means obtains the Mahalanobis general distance to the gene group constituting the data set excluding the training set and the Mahalanobis general distance to the gene group constituting the training set for the score calculation target gene, The apparatus according to claim 13, wherein the evaluation value is calculated as a difference value of the Mahalanobis general distance.   The apparatus according to claim 13, wherein the information related to the gene is disease type information.   The score calculation means is a score for one or more data selected from nucleotide sequence homology data, InterPro ID data, enzyme number data, metabolic pathway data, molecular function data in GO, biological process data, subcellular component data, and expression pattern data. The apparatus according to claim 13, wherein:   The score calculation means obtains the score for annotation data associated with a gene included in the training set, stores the score in a storage means as a table associated with the annotation data, and is associated with a score calculation target gene. The apparatus according to claim 13, wherein the table is searched by using the annotation data as a key, a score is read, and the read score is calculated as a score of annotation data related to the score calculation target gene.   14. The apparatus according to claim 13, wherein the information input via the input means or the communication network is a disease name.

Images