JP2005135154A - Method for predicting gene ontology term based on sequence similarity - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a method to make both recall values and accuracy values sufficiently higher than those in the conventional methods for the prediction of gene ontology terms based on sequence similarity. <P>SOLUTION: Using a gene sequence whose gene ontology terms are known, gene ontology terms are predicted and the accuracy of the prediction is calculated while the requirements for the prediction are varied. The optimal requirements for the prediction are searched and/or determined. Using the optimal requirements for the prediction, the gene ontology terms are predicted. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to information analysis of gene sequences, and relates to a gene ontology term prediction method for estimating a gene ontology term corresponding to a gene function by sequence similarity search.

  Conventionally, as a method for predicting a gene ontology term representing the characteristics of a gene sequence, there has been a method of extracting a functional motif possessed by a gene sequence and selecting a gene ontology term corresponding to the functional motif (Non-patent Document 1 below). . As a method based on sequence similarity, there is a method that uses a gene ontology term of a gene sequence that exceeds a certain threshold value as a prediction result using a gene sequence database that adopts gene ontology (the following non-patent document). Reference 2).

  However, the conventional method is not a method conscious of increasing the prediction accuracy, and the threshold value used in the conventional method is not a value determined to increase the prediction accuracy. There are two elements of prediction accuracy, the accuracy rate of prediction (Precision value) and the recovery rate of prediction (Recall value), and it is desirable that the Recall value and Precision value be sufficiently high.

Apweiler, R., et al., Nucleic Acids Res., 29: 37-40, 2001 The FANTOM Consortium and the RIKEN Genome Exploration Research Group Phase I & II Team, nature, 420: 563-573, 2002

  The problem to be solved by the present invention is to provide a method in which both the recall value and the precision value are sufficiently higher in gene ontology term prediction based on sequence similarity than in the conventional method.

  The present invention performs gene ontology term prediction by a method comprising the following processing steps in order to sufficiently increase both the recall value and precision value of gene ontology term prediction.

  That is, the gene ontology term prediction method according to the present invention provides a sequence similarity value between a first plurality of gene sequences to which a gene ontology term is assigned and a second plurality of gene sequences to which a gene ontology term is assigned. And calculating the sequence similarity value between the third gene sequence and the fourth plurality of gene sequences to which the gene ontology is assigned, and the step of determining a condition that the prediction accuracy of gene ontology prediction is sufficiently high. And assigning a gene ontology term to the third gene sequence according to the sequence similarity value and the conditions determined in the step.

  In order to prepare the first plurality of gene sequences and the second plurality of gene sequences, first, gene features in which gene features are already known and gene ontology terms representing the features are assigned to each entry It is preferable to determine the first plurality of gene sequences and the second plurality of gene sequences by using a database and dividing the database into two using random numbers.

  In addition, the gene ontology term prediction method according to the present invention provides a sequence similarity between each of a plurality of first gene sequences to which a gene ontology term is assigned and each of a plurality of second gene sequences to which a gene ontology term is assigned. Using the first step of calculating the sex value and the sequence similarity value calculated in the step, for each of the first plurality of gene sequences, the sequence similarity value with the first gene sequence is the first Selecting a gene sequence exceeding a threshold of 1 from the plurality of second gene sequences, a gene ontology term having an appearance frequency exceeding a second threshold among gene ontology terms assigned to the gene sequence selected in the previous period, A second step of predicting as a gene characteristic of the first gene sequence, and setting the first threshold and the second threshold to different values The first threshold value and the second threshold value are set for each of the third step, the second threshold value, and the second threshold value used in the second to third steps. Each of the prediction accuracy is obtained for the prediction results used, the first threshold value and the second threshold value corresponding to sufficiently high prediction accuracy among the predictions are determined, and the first threshold value is determined as the optimal sequence similarity threshold And a fourth step of setting the second threshold value as an optimum term appearance frequency threshold value, a third gene sequence, and each of a plurality of fourth gene sequences to which gene ontology terms are assigned as gene features A plurality of gene sequences each of which is obtained by obtaining a sequence similarity value, selecting a plurality of gene sequences whose sequence similarity value with the third gene sequence exceeds the optimum sequence similarity threshold, and Of gene ontology terms Frequency selects the Gene Ontology terms exceeds the optimal term occurrence frequency threshold, characterized in that it comprises a fifth step of predicting the gene ontology terms as a feature of a gene of the third gene sequence.

  The purpose of the first to fourth steps is to search for a sequence similarity threshold value and a term appearance frequency threshold value for predicting sufficiently high in prediction accuracy prior to actual prediction. Further, in order to obtain the prediction accuracy, the characteristics of an actual gene are known as the first sequence, and the first plurality of sequences to which gene ontology terms are assigned are used as the first sequence.

  In addition, the gene ontology term prediction method according to the present invention includes a first plurality of gene sequences, a second plurality of gene sequences, a third gene sequence, and a fourth plurality of gene sequences, wherein a nucleobase sequence or a protein amino acid sequence It is characterized by being.

  The gene sequences used in the present invention are preferably all nucleobase sequences or all protein amino acid sequences.

  The gene ontology prediction method according to the present invention is characterized in that a variable obtained from alignment between gene sequences or a function of the variable is used as the sequence similarity value.

  In the present invention, as a sequence similarity value, an E-value value used in a sequence homology search tool BLAST, a value corresponding to an E-value value used in a sequence homology search tool other than BLAST, an identity of alignment, or an alignment It is preferable to use the length or the ratio of the alignment length to the total length of both sequences to be aligned.

  The gene ontology term prediction method according to the present invention is not only a gene ontology term pre-assigned to the plurality of second gene sequences and the plurality of fourth gene sequences, but also a superordinate concept of the gene ontology term. Each gene ontology term is also characterized by calculating the appearance frequency in the previous term and making it a prediction target.

  The purpose of predicting a gene ontology term that is a superordinate concept of a gene ontology term pre-assigned to a gene sequence is to enable prediction of a gene ontology term of a superordinate concept and to increase the recall value of the prediction.

  In addition, the gene ontology term prediction method according to the present invention provides the sequence with the first gene sequence among a plurality of gene sequences in which the sequence homology value with the first sequence exceeds the first threshold. When n gene sequences having a relatively high homology value are selected and the total number of the same gene ontology terms included in the plurality of gene ontology terms assigned to the n gene sequences is m, m / n is defined as the appearance frequency of the gene ontology term.

  It is known that if the gene ontology term selected by the first threshold (sequence similarity value threshold) is used as a prediction result as it is, the Precision value will decrease because it contains many erroneous gene ontology terms. For each gene ontology selected with the first threshold, by calculating the appearance frequency and selecting a gene ontology with a high appearance frequency, it is assigned to a sequence with a higher sequence similarity value, and with a higher appearance frequency. The purpose is to select gene ontology terms that appear.

  The gene ontology term prediction method according to the present invention includes a first gene ontology term predicted for a gene sequence and a second gene ontology term previously assigned to the gene sequence, When one gene ontology term is located in a subordinate concept or a superordinate concept of the second gene ontology term, the second gene ontology term is a correctly predicted gene ontology term.

In the gene ontology term prediction method according to the present invention, the total number of gene ontology terms assigned in advance to the plurality of first gene sequences is A, and the gene ontology predicted for each of the plurality of first gene sequences. The total number of terms is B, the total number of correctly predicted gene ontology terms among the predicted gene ontology terms is C, the real number satisfying the following equation (1) is R, and the following equation (2) is satisfied When the real number to be performed is P, the function of the real number R and the real number P is used as the previous prediction accuracy.
R = C / A (1)
P = C / B (2)

  The real number R corresponds to the predicted recall value, and the real number P corresponds to the predicted precision value. Further, the prediction accuracy uses a function that becomes higher as the value of the real number R and the value of the real number P are both higher. Therefore, it is possible to search for a prediction condition in which the Recall value and the Precision value are both higher by searching for a prediction condition in which the prediction accuracy is higher.

  According to the present invention, gene ontology terms can be predicted with sufficiently high accuracy in gene ontology terms prediction based on sequence similarity search.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.
In the method of predicting the feature of the gene sequence given in FIG. 1 as a gene ontology term, it is intended to perform prediction with high prediction accuracy by obtaining an optimal prediction condition in advance. The flow of processing is shown. In the present embodiment, a case will be described in which protein amino acid sequences are used as gene sequences, and SWISS-PROT and TrEMBL published by SIB (Swiss Institute of Bioinformatics) on the Internet are used as protein databases.

  In FIG. 1, reference numeral 101 denotes a gene sequence that is a target for predicting a gene ontology term (hereinafter referred to as GO term for omission). Reference numeral 102 denotes gene sequence data in which gene features are known and GO terms corresponding to the features are assigned to entries. Reference numeral 103 denotes a file in which each entry of the 102 gene sequence database is associated with the GO term, and the file published by SIB on the Internet is used. 104 is a data file in which the GO terms are structured in a tree shape according to the hierarchical relationship of concepts, and the one published on the Internet by the Gene Ontology Consortium is used. There are three categories of molecular terms, biological processes, and cellular components in the GO terms developed by the Gene Ontology Consortium. Here, 104 files consist of molecular function terms. Next, 103 files are referenced, and entries having GO terms are searched from 102, and 105 databases comprising gene sequences having GO terms are created. At this time, in order to handle only GO terms assigned based on experimental evidence, 103 files are referenced, and the evidence code of GO terms is predicted by IEA (Inferred from electronic annotation) ) GO terms are preferably excluded. Next, the database 106 is divided into a test data set 107 and a learning data set 108. When five cross-validation tests are performed, the test data set 107 and the learning data set 108 are created by dividing the database 106 into five. At this time, it is preferable to randomly divide using random numbers. Next, in step 109, sequence similarity values between each gene sequence in the test data set 107 and each gene sequence in the learning data set 108 are calculated. For the calculation of the sequence similarity value, the program BLAST published by the US NCBI (National Center for Biotechnology Information) on the Internet or a similar program is used. At 110, while changing the prediction condition, a GO term is predicted for each prediction condition, and the prediction result is output to the gene sequence-known GO term-GO term correspondence file 111. In step 110, a gene ontology term assigned to the gene sequence of the learning data set 108 is selected that satisfies a certain prediction condition. In this step, not only the GO terms pre-assigned to 108 but also the GO terms located in all superordinate concepts of this GO term are predicted if they satisfy the prediction conditions. Selection of a GO term located in a superordinate concept of a certain GO term is performed with reference to the GO term tree structure file 104. The 111 files obtained in the 110 steps include (1) each gene sequence, (2) a known GO term pre-assigned to each gene sequence, and (3) a predicted GO term for each prediction condition. The relationship is described. Using the 111 files, 112 prediction accuracy calculation processing and optimum prediction conditions are determined. In this step, the prediction accuracy is calculated for each prediction condition used in 110, and the prediction condition (optimum prediction condition) that maximizes the prediction accuracy is determined. The steps so far are for obtaining the optimum prediction condition. Thereafter, GO term prediction is performed on the gene sequence 101 to be predicted using the optimum prediction conditions. In step 113, sequence similarity values are calculated between each of the 101 gene sequences to be subjected to GO term prediction and each of the 106 gene sequences in the database. In step 114, GO terms are assigned using the obtained sequence similarity value and the optimal prediction condition obtained in step 112, and the assigned GO terms are output to 115 as prediction results. In step 114, the GO term tree structure file 104 is used, and as in step 110, if the term of the higher concept satisfies the optimum prediction condition, it is output to 115 as a prediction result.

  FIG. 2 shows the data structure of data obtained from the result of the sequence similarity value calculation process 1 between the test data set array and the learning data set array in 109. 201 is data corresponding to the gene sequence of one test data set, and the whole data includes this repetitive structure. 201 is at least a name for identifying the gene sequence of the test data set, 202 GO term repetition structure pre-assigned to the test data set sequence, and learning data similar to the test data set sequence It contains a repetitive structure of information 203 on the gene sequence of the set. Reference numeral 202 denotes GO term information, which includes a GO term ID that identifies the GO term and the GO term. 203 includes at least the sequence name of the gene sequence of the learning data set and 204 repeat structures for the GO term to which the sequence is assigned. 204 is GO term information and includes at least an GO term or an ID for identifying the GO term.

  FIG. 3 is a flowchart for explaining the GO term assignment process at 110. The following processing is performed for all combinations of sequence similarity thresholds and term appearance frequency thresholds through repeated processing including 301 end determination. In 302, a combination of sequence similarity threshold and term appearance frequency threshold is set. The following processing is performed for all gene sequences in the test data set by repeated processing including 303 end determination. The information shown in 201 for the gene sequence of the test data set being processed in 303 is read. This includes a plurality of pieces of information on the gene sequence of the learning data set in which the sequence similarity shown in 202 is seen. In 305, the data whose sequence similarity value does not exceed the sequence similarity threshold is deleted from the plurality of pieces of data 203. However, when the E-value value of the BLAST tool is used as the sequence similarity value, the data is deleted when the sequence similarity value is equal to or less than the sequence similarity threshold. In 306, the appearance frequency is calculated for each of a plurality of 204 GO terms possessed by each of the 203 data. In 307, a GO term whose appearance frequency is equal to or higher than the appearance frequency threshold is selected. In 308, the selected GO term is set as the predicted GO term, and the test data set array name, the information on the predicted GO term, and the information on the 202 GO term are associated and output to the file 109.

  FIG. 4 shows the data structure of 109 files. 401 is data corresponding to the sequence similarity threshold and the term appearance frequency threshold set in 302, and the entire data has this repeating structure. 401 includes at least the sequence similarity threshold value set in 302, the term appearance frequency threshold data, and the repetition structure of the data 402 for all sequences in the test data set. 402 is at least the sequence name of the test data set and the data of the repeat structure 403 for the GO term predicted at 110 for the 402 sequence and the repeat structure 404 for the GO term pre-assigned to the 402 sequence. Contains data. However, if the GO term cannot be predicted for an array of a certain test data set, 402 does not include 403 data. 403 and 404 contain information on GO terms and / or GO term IDs.

  Next, the procedure of each process included in step 110 will be described using a specific example. These processes deal with 201 data, which has already been processed by 205 processes. For each gene sequence of the learning data set included in the 201 data, the GO terms and sequence similarity values possessed by the gene are extracted, and the table shown in FIG. 5 is created. For each GO term included in this table, all the terms (parent terms) located in the superordinate concept of the GO term are selected and added to the table of FIG. 5 to create the table shown in FIG. The parent term is selected by referring to the 104 GO term tree structure files. This file 104 describes the hierarchical relationship between the concepts of GO terms in a tree structure as shown in FIG. FIG. 7 shows the terms of the subordinate concept positioned in the direction of the arrow. The term B of 702 is a term (child term) positioned in the subordinate concept of the term A of 701. Next, referring to a table as shown in FIG. 6, GO terms that frequently appear in gene sequences having relatively high sequence similarity (small E-value values) are selected. For this purpose, attention is paid to the top N groups in descending order of sequence similarity (in the order of decreasing E-value values), and GO terms whose appearance frequency exceeds the appearance frequency threshold are selected. For example, GO term B appears at a frequency of 67% in the top 3 sequence groups of 701 (2 sequences out of 3). Thus, the appearance frequency of each term is calculated for each value of N from 1 to n (where n is the number of gene sequences selected by the sequence similarity threshold), and a table as shown in FIG. 8 is obtained. In step 307, the GO term whose appearance frequency is less than the term appearance frequency threshold is deleted from the table, and the remaining GO term is selected as the predicted GO term. When the term appearance frequency threshold is 70%, the GO terms A, B, D, and E are selected as the predicted GO terms with reference to the table of FIG. Considering the appearance frequency of the term in this way, even if the term is not assigned to the sequence having the highest sequence similarity value (the lowest E-value value), such as the term E, the appearance frequency is high overall. If it is a term, it can be selected as a predicted GO term. The predicted GO term selected in this way is processed into data having the data structure indicated by 402 and output, and after additional processing through 303 and 301, the file having the data structure shown in FIG. 111 is generated.

Next, details of 112 processes related to calculation of prediction accuracy and determination of optimal prediction accuracy will be described. All the information appearing in this process is obtained from data of 111 files having the data structure shown in FIG. A Recall value and a Precision value are calculated for each sequence homology threshold and term appearance frequency threshold contained in this file. The Recall value is the ratio of the total number of predicted GO terms out of the total number of GO terms in all the test data set arrays, and the ratio of the GO terms that could be predicted without missing out of the total GO terms to be predicted It is. The Precision value is the total number of correctly predicted GO terms in the total number of predicted GO terms, which means the prediction accuracy rate. Whether or not the GO term is correctly predicted is determined as follows. If the predicted GO term A and the GO term B pre-assigned to the test data set array are the same GO term, it is assumed that the GO term A is correctly predicted. Further, even if GO term A and GO term B are not the same, it is assumed that GO term B is correctly predicted as long as they are in a conceptual vertical relationship. Next, using the Recall value and Precision value,
(F-measure value) = 2 (Recall value) (Precision value) / ((Recall value) + (Precision value))
The F-measure value is calculated by using this value as the prediction accuracy. This F-measure value is an evaluation measure that increases as the Recall value and Precision value increase. In addition to the F-measure value, if there is another evaluation measure that increases as the Recall value and the Precision value become higher, the evaluation measure may be used as the prediction accuracy.

  The Recall value, Precision value, and F-measure value described above are calculated for each sequence similarity threshold and term appearance frequency threshold to obtain a table as shown in FIG. Referring to this table, the sequence similarity value and the term appearance frequency threshold that maximize the F-measure value are obtained, and the optimum sequence similarity value and the optimum term appearance frequency threshold are set as the optimal prediction. Condition. In addition, when 105 databases are divided into five and five cross-validation tests are performed, five sequence similarity values and term appearance frequency thresholds that maximize the F-measure value are obtained. It is preferable to use the average value as the optimum sequence similarity value and the optimum term appearance frequency threshold.

  Next, 113 to 115 processing procedures for predicting GO terms for 101 gene sequences using the optimal sequence similarity value and the optimal term appearance frequency threshold will be described. This procedure is basically the same as the procedure of outputting 111 files by the processing of 109 to 110. However, the GO term prediction target gene sequence 101 is used in place of the learning data set 108 used in the process 109, and 105 databases are used in place of the test data set 107. Furthermore, since the prediction using the optimal sequence similarity value and the optimal term appearance frequency threshold value is performed only once, the iterative process like 201 is not performed. By such processing, the GO term is predicted for each of the 101 sequences, and the result is output to 115 files.

  An attempt was made to determine the optimal sequence similarity threshold and the optimal term appearance frequency threshold using the above method. The protein amino acid sequence of SWISS-PROT, which is a public protein database, was used as the gene sequence data of 102 gene ontology terms. The 105 steps yielded 7830 sequences. Further, these sequences were divided into 1560 sequences by 5 and subjected to 5 cross-validation tests. The sequence similarity threshold was changed in steps from 1 to 1E-80, and the term appearance frequency threshold was changed in steps from 0% to 100%. The Recall value, Precision value, and F-measure value were calculated for the sequence similarity threshold and the term appearance frequency threshold, respectively. Since five cross-validation tests were performed, the average value of the five tests was used for the Recall value and Precision value. As a result, when the sequence similarity threshold was 0.01 and the term appearance frequency threshold was 60%, the F-measure value was 0.63, which was the maximum. The Recall value and Precision value at this time were 0.55 and 0.75, respectively. For comparison, a sequence similarity threshold 1E-10, which is often used for gene function prediction, was used for prediction without selection based on the appearance frequency of terms. As a result, the Recall value was 0.60, the Precision value was 0.39, and the F-measure value was 0.47. Therefore, the prediction accuracy of this method exceeded the comparison method in terms of F-measure value, and the effectiveness of this method was proved.

Hereinafter, the procedure of the present invention will be described with reference to FIG.
<Procedure 1>
The GO term known gene sequence data 1011 and the gene sequence-GO term correspondence data 1012 are inputted, and a program for creating a database of 1005 GO term known gene sequences is outputted as 1014 GO term known gene sequence database.

<Procedure 2>
A program for creating a database of 1006 GO term known gene sequences is used to input a 1014 GO term known gene sequence database and output a learning data set 1015 and a test data set 1016.

<Procedure 3>
The training data set 1015 and the test data set 1016 are input by the program that calculates the sequence similarity value between the gene sequence of the training data set 1006 and the gene sequence of the test data set. Output sequence similarity value data with a set of gene sequences.

<Procedure 4>
Using a program that predicts the GO term of the gene sequence of the learning data set with multiple prediction conditions of 1007, the sequence similarity value data of the gene sequence of the 1017 learning data set and the gene sequence of the test data set, and 1013 The GO term tree structure file is input, and the correspondence data between the known GO term of the gene sequence of the learning data set and the predicted GO term in each of 1018 prediction conditions is output.

<Procedure 5>
Using the program that calculates the prediction accuracy under each prediction condition of 1008 and determines the optimal prediction condition, input the correspondence data of the known GO term of the gene sequence of the learning data set and the predicted GO term in each prediction condition of 1018 Determine optimal prediction conditions.

<Procedure 6>
Using a program for determining sequence similarity values between 1009 GO term predicted gene sequences and GO term known gene sequences, 1019 GO term predicted gene sequences and 1014 GO term known gene sequence database were input, 1020 The sequence similarity value data between the GO term prediction target gene and the GO term known gene sequence is output.

<Procedure 7>
Using a program that predicts the GO term against the target sequence of GO terms based on the optimal prediction conditions of 1010, the sequence similarity value data of 1020 GO terms predicted gene sequence and the known GO gene sequence, and 1013 GO Using the term tree structure file, output 1021 GO term prediction results.

<Procedure 8>
Using 1002 display and 1003 pointing device, output 1021GO term prediction results on 1002 display.

  Although the procedure is as described above, the data to be stored in advance in the auxiliary storage device is at least GO term known gene sequence data 1011, gene sequence-GO term correspondence data 1012, GO term tree structure file 1013, GO term. This is the predicted gene sequence 1019. The remaining data of the other auxiliary storage device can be generated in the calculation process.

The flowchart for demonstrating the whole flow of the gene ontology term prediction method of this invention. Data structure of data obtained from the result of sequence similarity value calculation. The flowchart for demonstrating the flow of a gene ontology term allocation process. The data structure of the data used to calculate the prediction accuracy. A table representing data relating to sequences selected by sequence similarity thresholds. The table | surface showing the data by which the gene ontology term of the high-order concept was added to the gene ontology term of the arrangement | sequence selected by the sequence similarity threshold value. Explanatory drawing showing the tree structure of gene ontology. A table representing data obtained by searching for an optimal prediction condition. The table which calculated the Recall value, Precision value, and F-measure value for every sequence similarity threshold value and term appearance frequency threshold value. The system block diagram explaining the procedure of this invention.

Explanation of symbols

101: Gene sequence data to be predicted, 102: Gene sequence data with known gene ontology terms, 103: Corresponding file of gene sequence names with known gene ontology terms and gene ontology terms, 104: Tree of gene ontology terms Structure file, 105: processing for creating a database regarding gene sequences whose gene ontology terms are known for sequence similarity value calculation processing, 106: gene sequence database for known gene ontology terms used for sequence similarity value calculation processing, 107: gene sequence test data set used for searching for optimal prediction conditions, 108: gene sequence learning data set used for searching for optimal prediction conditions, 109: sequence similarity value calculation processing, 110: gene ontology term Assignment processing, 111: Gene ontology assignment 112: calculation of prediction accuracy and determination processing of optimum prediction conditions, 113: calculation processing of sequence similarity values between gene ontology prediction target sequences and sequences of gene ontology terms known gene sequence database, 114: Gene ontology term prediction process data assignment process to gene ontology term prediction target gene sequence, 115: gene ontology term prediction result data file.
202: Data on gene ontology terms pre-assigned to sequences in test data set, 203: Data on learning data set with sequence similarity with test data set, 204: Data with sequence similarity with test data set Data related to gene ontology terms that are pre-assigned to different learning data sets.
403: Data related to gene ontology terms predicted for the sequence of the test data set, 404: Data related to gene ontology terms pre-assigned to the sequences of the test data set.
601: The top three sequence groups.
701-702: A gene ontology term included in the tree structure of a gene ontology.

Claims (8)

Using the sequence similarity value between the first plurality of gene sequences to which the gene ontology term is assigned and the second plurality of gene sequences to which the gene ontology term is assigned, the prediction accuracy of gene ontology prediction is sufficiently high A step of determining a condition to increase;
A sequence similarity value between a third gene sequence and a fourth plurality of gene sequences assigned with gene ontology is calculated, and the third gene is calculated according to the sequence similarity value and the conditions determined in the step. Assigning gene ontology terms to sequences. A method for predicting gene ontology terms. Calculating a sequence similarity value between each of the first plurality of gene sequences assigned to the gene ontology term and each of the second plurality of gene sequences assigned to the gene ontology term;
Using the sequence similarity value calculated in the step, for each of the first plurality of gene sequences, a gene sequence having a sequence similarity value with the first gene sequence that exceeds a first threshold is the second A gene ontology term having an appearance frequency exceeding a second threshold value among gene ontology terms assigned to the gene sequence selected in the previous period is selected as a gene characteristic of the first gene sequence. A second step to predict;
A third step of repeating the second step while setting the first threshold and the second threshold to different values;
For each of the first threshold value and the second threshold value used in the second to third steps, a prediction accuracy is obtained for each prediction result using the first threshold value and the second threshold value, and the prediction Among the first threshold value and the second threshold value corresponding to sufficiently high prediction accuracy, the first threshold value as an optimal sequence similarity threshold value, and the second threshold value as an optimal term appearance frequency threshold value. And a fourth step,
A sequence similarity value between each of the third gene sequence and each of a plurality of fourth gene sequences to which gene ontology terms are assigned as gene characteristics is obtained, and the sequence similarity value with the third gene sequence is determined. Selects a plurality of gene sequences exceeding the optimum sequence similarity threshold, and among gene ontology terms assigned to the plurality of gene sequences, a gene ontology whose appearance frequency exceeds the optimum term appearance frequency threshold A gene ontology term prediction method comprising: selecting a term and predicting the gene ontology term as a gene characteristic of the third gene sequence.   3. The gene ontology term prediction method according to claim 2, wherein the first plurality of gene sequences, the second plurality of gene sequences, the third gene sequence, and the fourth plurality of gene sequences are a nucleobase sequence or a protein amino acid sequence. A gene ontology term prediction method characterized by   The gene ontology term prediction method according to any one of claims 2 to 3, wherein a variable obtained from alignment between gene sequences or a function of the variable is used as the sequence similarity value. Method.   In the gene ontology term prediction method according to any one of claims 2 to 4, not only the gene ontology terms pre-assigned to the plurality of second gene sequences and the plurality of fourth gene sequences, A gene ontology term prediction method, characterized by calculating the appearance frequency of each gene ontology term, which is a superordinate concept of the gene ontology term, as a prediction target.   The gene ontology term prediction method according to any one of claims 2 to 5, wherein, among the plurality of gene sequences whose sequence similarity value with the first sequence exceeds the first threshold, the first The total number of the same gene ontology terms included in the plurality of gene ontology terms selected from the n gene sequences having a relatively high sequence homology value with the gene sequence of A gene ontology term prediction method, wherein m / n is an early appearance frequency of the gene ontology term, where m is n.   The gene ontology term prediction method according to any one of claims 2 to 6, wherein the first gene ontology term predicted for a certain gene sequence is the same as the second gene ontology term previously assigned to the gene sequence. And when the first gene ontology term is located in a subordinate concept or a superordinate concept of the second gene ontology term, the second gene ontology term is a correctly predicted gene ontology term A gene ontology term prediction method characterized by that. The gene ontology term prediction method according to any one of claims 2 to 7, wherein A is a total number of gene ontology terms preassigned to the plurality of first gene sequences, and the plurality of first gene sequences. The total number of gene ontology terms predicted for each is B, the total number of correctly predicted gene ontology terms among the predicted gene ontology terms is C, and the real number satisfying the following equation (1) is R: A gene ontology term prediction method, wherein a real number satisfying the formula (2) is P, and a function of the real number R and the real number P is used as a prediction accuracy.
R = C / A (1)
P = C / B (2)

Images