program

Abstract

The program instructs a computer to perform an FPO analysis process for each rank while increasing the rank, in which a seed element module is constructed from non-seed elements with high correlations with the seed element among elements contained in multivariate data that are the subject of correlation network analysis. The FPO analysis process removes false positive non-seed elements from among the non-seed elements with high correlations with the seed element, stores configuration information of the module on which the FPO analysis process has been performed in a memory unit, and if a module identical to the module configuration information stored in the memory unit is constructed, stops the FPO analysis process and early terminates the search for the module with the highest module F-score for that rank. The program instructs the computer to perform an FNI analysis process that incorporates false negative elements, and, simultaneously with the FNI analysis process, calculates the element specificity rate for each correlation rank of candidate elements that are outside the module with respect to the element of interest, while calculating an element P-value using a recursive formula.

Description

[Technical Field]

[0001] The present invention relates to a program that causes a computer to analyze the relationships between multiple components that influence each other, identify modules that have a high impact on a particular effect, and extract candidates. [Background technology]

[0002] Patent Document 1 describes a technique for performing correlation network analysis on multivariate data, which is comprehensive molecular information obtained by omics analysis. Non-Patent Document 1 describes a technique for performing correlation network analysis on multivariate data such as big data. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 6318334 [Non-patent literature]

[0004] [Non-Patent Document 1] Kazuto Mannen, Yoshiyuki Ogata, Hideyuki Suzuki, "A breakthrough in multivariate analysis - Development of ConfeitoGUIplus", Journal of Bioengineering, Vol. 95, No. 7, July 2017 Summary of the Invention [Problem to be solved by the invention]

[0005] However, if the techniques described in Patent Document 1 and Non-Patent Document 1 are implemented as they are, the amount of calculation required will be extremely large, making their application to big data impractical.

[0006] An object of the present invention is to provide a program that can reduce the amount of calculation required for correlation network analysis of multivariate data. [Means for solving the problem]

[0007] One aspect of the present invention is a program that causes a computer to perform an FPO analysis process for each rank, increasing the rank from a lower limit to an upper limit, which is the number of elements other than the seed elements to be incorporated into the seed element module, when constructing a seed element module from non-seed elements that have a high correlation with the seed element among elements contained in multivariate data that are the subject of correlation network analysis. The FPO analysis process removes false positive non-seed elements from among the non-seed elements that have a high correlation with the seed element, and stores configuration information of the module on which the FPO analysis process has been performed in a memory unit.If a module identical to the configuration information of the module stored in the memory unit is constructed, the FPO analysis process is stopped and the search for the module that has the highest module F-score of that rank is terminated early.

[0008] In one aspect of the present invention, a computer is caused to execute an FNI analysis process for incorporating false negative elements into a plurality of modules composed of elements contained in multivariate data that are the subject of correlation network analysis, and simultaneously with the FNI analysis process, the computer is caused to calculate an element specificity rate for each correlation order of candidate elements that are elements outside the module with respect to the element of interest, while calculating an element P value using the following formula:

number

[0009] According to the present invention, it is possible to provide a program that can reduce the amount of calculation required for correlation network analysis of multivariate data. [Brief explanation of the drawings]

[0010] [Figure 1] 1 is a diagram illustrating an example of a schematic configuration of an information processing device according to an embodiment; [Figure 2]1 is a flowchart illustrating an example of a procedure of an information processing method according to an embodiment. [Figure 3] FIG. 1 is a diagram illustrating an example of the configuration of multivariate data according to an embodiment. [Figure 4A] FIG. 10 is a diagram illustrating an example of a correlation coefficient between elements according to an embodiment. [Figure 4B] FIG. 10 is a diagram illustrating an example of a correlation coefficient between elements according to an embodiment. [Figure 5A] 10A and 10B are diagrams for explaining an FPO module selection method according to an embodiment. [Figure 5B] 10A and 10B are diagrams for explaining an FPO module selection method according to an embodiment. [Figure 5C] FIG. 10 is a diagram for explaining an example of FPO analysis according to an embodiment. [Figure 6A] FIG. 10 is a diagram illustrating an example of a final element F-score according to an embodiment. [Figure 6B] FIG. 10 is a diagram illustrating an example of a final element F-score according to an embodiment. [Figure 6C] FIG. 10 is a diagram illustrating an example of a method for calculating a final element F-score according to an embodiment. [Figure 7A] FIG. 10 is a diagram illustrating an example of integration processing according to an embodiment. [Figure 7B] FIG. 10 is a diagram illustrating an example of integration processing according to an embodiment. [Figure 8] FIG. 10 illustrates an example of a final FPO module integration result according to one embodiment. [Figure 9A] 9 is a diagram showing an example of an integration result of FPO modules corresponding to the final integration result of FPO modules in FIG. 8. FIG. [Figure 9B] 9 is a diagram for explaining an example of calculation of a distance index of the final FPO module of FIG. 8. FIG. [Figure 10] FIG. 10 is a diagram for explaining an FNI analysis process according to an embodiment. [Figure 11] FIG. 10 is a diagram showing an example of the results of comparing the analysis time of element P values ​​with and without a recursive formula. DETAILED DESCRIPTION OF THE INVENTION

[0011] Next, the program of this embodiment will be described with reference to the drawings. The embodiment described below is merely an example, and the embodiments to which the present invention is applied are not limited to the following embodiments. In all drawings used to explain the embodiment, the same reference numerals are used for components having the same functions, and repeated explanations will be omitted.

[0012] Furthermore, in this application, "based on XX" means "based on at least XX," and includes cases where it is based on other elements in addition to XX. Furthermore, "based on XX" is not limited to cases where XX is used directly, but also includes cases where it is based on XX that has been calculated or processed. "XX" is any element (for example, any information).

[0013] The terms used in this embodiment are explained in Tables 1, 2 and 3 below.

[0014] [Table 1]

[0015] [Table 2]

[0016] [Table 3]

[0017] 1 is a diagram showing an example of a schematic configuration of an information processing device 1 according to this embodiment. In FIG. 1, the information processing device 1 includes an input / output unit 11, a storage unit 12, and a control unit 13.

[0018] The input / output unit 11 receives input of data and outputs data to and from devices external to the information processing device 1. The method of inputting and outputting data is not limited. The input / output unit 11 may transmit and receive data by communication via, for example, a communication line. The input / output unit 11 may input and output data via, for example, a recording medium.

[0019] The storage unit 12 stores various data such as programs executed by the control unit 13 and data used by the control unit 13. The storage unit 12 is configured by a non-volatile memory such as a hard disk device, a magneto-optical disk device, or a flash memory, a read-only recording medium such as a CD-ROM, a volatile memory such as a DRAM (Dynamic Random Access Memory), or a combination of these.

[0020] The control unit 13 includes a CPU (Central Processing Unit) and realizes various functions by executing programs stored in the storage unit 12. When the control unit 13 according to this embodiment executes the correlation network analysis program 121 stored in the storage unit 12, the functions of a correlation coefficient calculation unit 131, an FPO analysis unit 132, an FPO module selection unit 133, an FPO module integration unit 134, an FNI analysis unit 135, an FNI assignment processing unit 136, a P value calculation unit 137, and an output data generation unit 138 are realized.

[0021] The information processing device 1 may be configured using a general-purpose computer device, or may be configured as a dedicated hardware device. For example, the information processing device 1 may be configured using a server computer connected to a communication network such as the Internet. Furthermore, each function of the information processing device 1 may be realized by cloud computing. Furthermore, the information processing device 1 may be realized by a single computer, or may be realized by distributing the functions of the information processing device 1 among multiple computers. Furthermore, the information processing device 1 may be configured to open a website using, for example, a WWW system.

[0022] The storage unit 12 has storage areas for storing a correlation coefficient list 122, module information for each rank by element 123, module integration information for each module F score (MF) and element F score (VF) threshold 124, final module integration information 125, evaluation value information for each target step 126, final module information 127, and an FPO analyzed module list 128. Note that a hash value may be calculated and added to each piece of information to speed up searches of the information stored in the storage unit 12.

[0023] Next, an information processing method according to this embodiment will be described with reference to Fig. 2. Fig. 2 is a flowchart showing an example of the procedure of the information processing method according to this embodiment.

[0024] (Step S1) The input / output unit 11 accepts input data. The input data is multivariate data on which correlation network analysis is to be performed. The input data may be, for example, comprehensive molecular information obtained by omics analysis. The input data may be, for example, information about a product obtained in the manufacturing and inspection processes of the product for the purpose of analyzing the cause of defective products. The input data may be, for example, big data. The input data may be, for example, time-series data.

[0025] The multivariate data that is the subject of the correlation network analysis according to this embodiment is not particularly limited in terms of its acquisition route, acquisition method, type, attributes, etc. For example, multivariate data obtained by omics analysis can be cited. "Omics analysis" generally refers to the integrated analysis of comprehensive molecular information on individual molecules, and representative examples of comprehensive molecular information include transcriptome data, which is comprehensive information on gene transcripts, and metabolome data, which is comprehensive information on metabolites.

[0026] These comprehensive data can be obtained by any method or means, such as various genetic analyses, gene expression analyses, and various mass analyses such as liquid chromatography-mass spectrometry (LC-MS), gas chromatography-mass spectrometry (GC-MS), and capillary electrophoresis-mass spectrometry (CE-MS). Furthermore, the source of this comprehensive data is not particularly limited and can include parts, organs, tissues, and cells derived from various types of animals, plants, microorganisms, and bacteria. Furthermore, multivariate data can be information obtained by any method from samples obtained from a certain environment and artificially manufactured products (e.g., processed foods).

[0027] The input data may be in the form of a plurality of data sets each consisting of a plurality of quantitative values ​​corresponding to each of a plurality of elements. FIG. 3 is a diagram showing an example of the configuration of multivariate data (input data) according to this embodiment. As shown in FIG. 3, the input data may be a quantitative value table consisting of rows corresponding to each element and columns corresponding to each data set. In the quantitative value table shown in FIG. 3, the first column is the element name (e.g., element identifier (ID001, ID002, etc. in FIG. 3)), the first row is the data set name (DATA_a, DATA_b, etc. in FIG. 3), and the quantitative value of data set y for element x is stored in row x and column y (x and y are integers equal to or greater than 2). For example, the data "data_b002" in row 3 and column 3 is the quantitative value of data set "DATA_b" for element "ID002".

[0028] (Step S2) The correlation coefficient calculation unit 131 calculates the correlation coefficient between elements based on the quantitative values ​​of the input data. The correlation coefficient calculation unit 131 generates a correlation coefficient list 122 by sorting the correlation coefficients for each element in the input data in descending order. In this way, correlation coefficient lists 122 are generated for all elements in the input data. The correlation coefficient list 122 for element x is a list in which pairs of element names and correlation coefficients are sorted in descending order of the correlation coefficient with element x. The correlation coefficient lists 122 for each element generated by the correlation coefficient list 122 are stored in the storage unit 12.

[0029] The correlation coefficient between elements according to this embodiment is, for example, one of the following indices: Pearson's product-moment correlation coefficient, Spearman's rank correlation coefficient, and cosine similarity. The correlation coefficient between elements is a real number between -1 and 1.

[0030] In addition, as for the correlation coefficient between elements according to this embodiment, negative correlation coefficients may be replaced with "0", or they may not be replaced with "0". When negative correlation coefficients are not replaced with "0", it is possible to distinguish between "negative correlation" and "no correlation". Therefore, even when two elements that have a strong positive correlation with another element and a negative correlation belong to the same module, they can be matched. This makes it easier to interpret the modules.

[0031] 4A and 4B are hypothetical image diagrams for explaining an example of a correlation coefficient between elements according to this embodiment. FIG. 4A shows an example of a module, with elements indicated by circles. Elements with a correlation coefficient equal to or greater than a specific value are connected by lines. The elements include seed elements and elements within the module. Symbols shown in the circles are element identification information. In the example shown in FIG. 4A, the correlation coefficient ρ between element X and element Y is XY is greater than zero, and the correlation coefficient ρ between element X and element Z XZ Consider the case where is greater than zero. In this case, the correlation coefficient ρ between element Y and element Z YZ It is unclear whether or not is greater than zero. Figure 4B shows the correlation coefficient ρ YZ The following figure shows the possible values ​​of the correlation coefficient ρ YZ It can be seen that ρ can be greater than zero, less than zero, or it can be unknown whether it is greater than or less than zero. ρ denotes the correlation coefficient. Figure 4B shows the range of Pearson correlation values ​​(-1 to 1) on the X and Y axes.

[0032] (Step S3) The FPO analysis unit 132 sets all elements of the input data as seed elements, and executes a process (FPO analysis process) to exclude (Out) false-positive non-seed elements from among the non-seed elements positioned at the top (those with larger correlation coefficients) in the correlation coefficient list 122 of the seed elements. The FPO analysis process will be described below.

[0033] The FPO analysis unit 132 constructs a provisional module by connecting a predetermined number of top (up to Mth) non-seed elements in the correlation coefficient list 122 of the seed element with edges. M is a value within a predetermined range of module sizes (from the lower limit to the upper limit of the module size). The number of elements (excluding the seed element) that constitute the module of the seed element is the rank (rank = M). The predetermined range of module size can be arbitrarily set by the operator, for example, by setting a MIN value and a MAX value in the information processing device 1 in advance. Here, the MIN value and MAX value are the module size excluding the seed element and are the minimum and maximum values ​​that the rank can take. Therefore, the lower limit of the total number of elements (seed elements and non-seed elements) that constitute the module (module size) is "MIN value + 1", and the upper limit is "MAX value + 1".

[0034] Next, the FPO analysis unit 132 generates a list for the provisional module in which the correlation coefficients between all elements included in the correlation coefficient list 122 for the seed element and each non-seed element are sorted in descending order. The FPO analysis unit 132 further connects elements in the generated list with edges whose correlation coefficients are greater than a predetermined correlation threshold. By setting multiple correlation thresholds, multiple provisional modules with different edge patterns are constructed. For example, by setting all possible correlation threshold values ​​as the correlation threshold, provisional modules with edge patterns for all of the correlation thresholds are constructed. The lower the correlation threshold, the greater the number of edges. For example, instead of simultaneously setting multiple correlation thresholds, the correlation threshold may be initialized to the maximum value "1" and then reduced while constructing provisional modules with each correlation threshold. Next, the FPO analysis unit 132 calculates the F score and MF of each of the provisional modules having different edge patterns using the following equations.

[0035]

number

[0036] n is the number of elements in the module. edge(i) is the number of edges of element i, which is the number of elements connected to element i within the same module for the edge pattern corresponding to the current correlation coefficient. i is an integer value between 1 and n, and is the index of the element. degree(i) is the degree of element i, which is the total number of elements connected to element i both inside and outside the module for the edge pattern corresponding to the current correlation coefficient.

[0037] MF is the harmonic mean of module density (MD) and module specificity (MS). MD represents the degree to which a module is internally connected as a whole. MS represents the degree to which a module is isolated from the outside (exclusively connected). Therefore, a module with a large MF can be said to be isolated from the outside and have strong connections as a whole.

[0038] The FPO analysis unit 132 stores in the storage unit 12 the maximum value of the MFs calculated for each of a plurality of provisional modules with different edge patterns, and the correlation threshold value and edge pattern corresponding to the maximum value of the MF. The FPO analysis unit 132 calculates the correlation threshold corresponding to the maximum value of MF and the F score, VF(i), of each non-seed element i for the edge pattern using the following equation.

[0039]

number

[0040] VF(i) is the harmonic mean of element density (VD(i)) and element singularity (VS(i)). VD(i) represents the proportion of element i connected to other elements within the module. VS(i) represents the proportion of the degree element i has relative to the entire network that is connected to elements of the module. i is an integer value between 1 and n, and is the index of the element. Therefore, an element i with a large VF(i) is closely and exclusively connected to the module, and can be said to be an element suitable as a component of the module.

[0041] The FPO analysis unit 132 determines the non-seed element i with the smallest VF(i) of the provisional module as a false positive and removes the non-seed element i determined to be a false positive from the provisional module. Next, the FPO analysis unit 132 recalculates the maximum MF and VF(i) of each non-seed element i for the provisional module from which the non-seed element i determined to be a false positive has been removed. This FPO analysis process of removing false positive non-seed elements i is repeated until the number of elements in the provisional module reaches the "module size lower limit," or "MIN value + 1." The FPO analysis unit 132 determines that the point at which the MF is maximum is the state in which no false positives have been removed, and stores the configuration information of the provisional module at the time when the MF is maximum in the module information 123 as the maximum MF module in the "rank" of the seed element.

[0042] The FPO analysis unit 132 executes FPO analysis processing for each rank within a predetermined range of ranks (from the lower limit (MIN value + 1) to the upper limit (MAX value + 1) of the module size at the time the provisional module is constructed). As a result, for one seed element, for each rank within the predetermined range of ranks, the configuration information of the provisionally assembled module at the time when the MF was maximum is stored in the module information 123 as the maximum MF module.

[0043] The FPO analysis unit 132 performs the FPO analysis in step S3 while increasing the rank from the lower limit (MIN value) of the rank to the upper limit (MAX value), and stores the configuration information of the module (FPO-analyzed module) for which the FPO analysis in step S3 has been completed in the FPO-analyzed module list 128. If a module identical to an FPO-analyzed module already existing in the FPO-analyzed module list 128 is constructed, the FPO analysis process for removing false-positive non-seed elements may be stopped and the search for the maximum MF module of that rank may be terminated early.

[0044] The FPO analysis unit 132 determines whether the maximum value of the MF is not updated using the correlation threshold and the edge pattern, and if it determines that the maximum value of the MF is not updated for a temporary module whose MF has not been calculated among multiple temporary modules with different edge patterns, it may terminate the calculation of the MF and early terminate the search for the maximum MF module of that rank. By configuring in this manner, when the FPO analysis unit 132 determines that the maximum MF value will not be updated for a temporary module for which the MF has not been calculated among multiple temporary modules with different edge patterns, it can terminate the MF calculation and early terminate the search for the maximum MF module of that rank.Therefore, even if the maximum MF value is not updated, the amount of calculation can be reduced and the FPO analysis processing time can be shortened compared to when the MF calculation continues and the module optimization processing continues.

[0045] (Step S4) The FPO module selection unit 133 selects an optimal module (referred to as an FPO module) based on the module information 123 for a certain seed element as a result of the FPO analysis.

[0046] FIG. 5A is a diagram illustrating the FPO module selection method according to this embodiment. The example in FIG. 5A is time-series metabolome data of grated radish (a dataset of 2305 elements x 10 time-series points). The example shows the results of FPO analysis using an element with identifier "c1487" as the seed element, with a Pearson product-moment correlation coefficient, a MIN value of "2," and a MAX value of "2305." FIG. 5A shows the maximum MF value (vertical axis) corresponding to the rank (horizontal axis). The FPO module selection unit 133 determines the module M_PTmax of the rank that indicates the maximum PTmax (longest plateau length) during the period in which the maximum MF value is not updated even when the rank value is increased. The FPO module selection unit 133 selects the determined module M_PTmax as the FPO module.

[0047] In the example shown in Figure 5A, the MF of the module constructed at "Rank = 623" was calculated to be 0.683, and no modules with an MF exceeding 0.683 were constructed up to "Rank = 761." Therefore, the plateau length, which is the period during which the maximum MF is not updated even when the rank value is increased, is "138," obtained by subtracting "Rank = 623" from "Rank = 761." In this example, "138" is the longest plateau length (longest plateau length) PTmax, and therefore the module constructed at "Rank = 623" was selected as the module M_PTmax of rank "623," which exhibits the longest plateau length PTmax.

[0048] Previously, the module with the rank showing the maximum MF across all ranks was selected as the FPO module. However, the rank showing the maximum MF across all ranks is often at or near the upper limit (MAX value) of the rank. For this reason, previously, large, unbalanced modules were often configured. Furthermore, previously, because it depended on the setting of the upper limit (MAX value) of the rank, there was a possibility that optimal FPO analysis results could not be obtained.

[0049] In contrast, according to this embodiment, a module with a large MF and an appropriate number of elements can be obtained, thereby improving the quality of the FPO analysis results. Furthermore, the FPO analysis results are less affected by the setting of the upper limit value (MAX value) of the rank. This allows for a more robust FPO analysis.

[0050] Furthermore, the module information 123 for each rank can be used to verify the selection criteria for the FPO module. The output data generation unit 138 may use the module information 123 for each rank to generate data for a graph of the maximum MF value, as exemplified in FIG. 5A. This graph data is data that can be used to display the graph on a display screen or print the graph. This allows the operator to easily verify the selection criteria for the FPO module.

[0051] Alternatively, other FPO module selection methods may employ criteria that maintain robustness regardless of the number of elements in the module or the distribution of correlation coefficients between elements. For example, criteria using the plateau length of the graph of the moving average of the maximum MF values, as illustrated in Figure 5B, or criteria that the increase rate of the graph is below a certain value may be employed. Furthermore, criteria optimized for each data set based on the distribution of correlation coefficients between elements may also be employed. In this case, the upper limit of the rank, i.e., the MAX value, does not need to be adjusted as a setting parameter, thereby reducing the burden on the operator.

[0052] FIG. 5B is a diagram illustrating the FPO module selection method according to this embodiment. The example in FIG. 5B is time-series metabolome data for grated radish (a 10-point time-series dataset with 2305 elements). The results show the results of FPO analysis performed using an element with identifier "c1487" as the seed element, with a Pearson product-moment correlation coefficient, a MIN value of "2," and a MAX value of "2305." FIG. 5B shows a 10-point moving average (vertical axis) of the maximum MF corresponding to the rank (horizontal axis). In the example shown in FIG. 5B, the MF of the module constructed at "Rank = 152" was calculated to be 0.590, and no modules with an MF exceeding 0.590 were constructed up to "Rank = 385." Therefore, the plateau length of the moving average, which is the period during which the maximum MF is not updated even when the rank value is increased, is "233," calculated by subtracting "Rank = 152" from "Rank = 385." In this example, "233" was the longest plateau length (longest plateau length) MAPTmax of the moving average, so the module constructed with "rank = 152" was selected as the module M_MAPTmax with rank "152" which indicates the longest plateau length MAPTmax.

[0053] The FPO module selection unit 133 may terminate the FPO analysis and the selection of the FPO module when a predetermined condition is satisfied. FIG. 5C is a diagram illustrating an example of the FPO analysis according to this embodiment. FIG. 5C shows time-series metabolome data of grated radish (a dataset of 2305 elements and 10 time-series points). The results show the results of FPO analysis performed using an element with identifier "c1487" as a seed element, with a Pearson product-moment correlation coefficient of MIN value "2" and MAX value "2305." FIG. 5C shows the maximum MF (vertical axis) corresponding to the rank (horizontal axis). In this example, since there are 1569 elements with a positive correlation coefficient with identifier "c1487," the maximum rank of identifier "c1487" is 1569. FIG. 5C illustrates the process performed in step S3 when the FPO module selection unit 133 is executed and determines whether the longest plateau length is updated each time the FPO analysis unit 132 finishes calculating the maximum MF for each rank. When the FPO module selection unit 133 determines that the longest plateau length has not been updated, it may determine that the maximum value of the plateau length has not been updated even if the rank changes (even if the rank increases), and may terminate the increase in rank and terminate the FPO analysis and selection of the FPO module. When the rank exceeds 1432 in FIG. 5C, the remaining number of ranks is "137", which is the maximum rank "1569" minus the rank "1432". Since the number of ranks "137" is below the PTmax "138", it is determined that the portion of the rank equal to or greater than 1432 does not exceed PTmax, and calculation is not performed for the portion marked "no calculation required" (portion of the rank equal to or greater than 1432). By configuring in this manner, the FPO module selection unit 133 can terminate FPO analysis and selection of FPO modules if the longest plateau length is not updated even when an element is added (even when the rank is increased).Therefore, even if the longest plateau length is not updated, the amount of calculation can be reduced and the time required for FPO analysis and selection of FPO modules can be shortened compared to when FPO analysis and selection of FPO modules are continued for each rank.

[0054] (Step S5) If the selection of FPO modules has been completed using all elements in the input data as seed elements (YES in step S5), proceed to step S6-1. On the other hand, if there are still elements for which an FPO module has not been selected (NO in step S5), return to step S3 and perform FPO analysis and FPO module selection using the elements for which an FPO module has not yet been selected as seed elements.

[0055] (Step S6-1) The FPO analysis unit 132 calculates, for each seed element's FPO module, the maximum value of VF(i) for each element (i) in the FPO module (referred to as the final VF) and an MF (referred to as the final MF) based on the final VF. The final MF and final VF calculated for each seed element's FPO module are stored in the evaluation value information 126.

[0056] 6A and 6B are hypothetical conceptual diagrams for explaining an example of a final VF according to this embodiment. FIG. 6A shows an example of a module and elements. In FIG. 6A, elements are indicated by circles. The elements include a target element (element for which the final VF is calculated), elements within the module, and elements outside the module. The numerical values ​​indicated in the circles are correlation rankings for the target element. In this example, the higher the correlation, the higher the correlation ranking (the smaller the value). In FIG. 6A, lines connect elements with correlation coefficients equal to or greater than a specific value between the target element and elements within the module, and between the target element and elements outside the module. Solid lines indicate elements included in edges, i.e., relationships between elements within the module. Dotted lines indicate elements not included in edges, i.e., relationships between elements outside the module.

[0057] For one FPO module, the FPO analysis unit 132 calculates a final VF for each element (i) in the FPO module. Specifically, the FPO analysis unit 132 sets each element (i) in the FPO module as an element of interest and calculates a VF(i) for the element of interest. FIG. 6B shows the calculation result of VF(i) for one element (i) (element of interest). In FIG. 6B, the horizontal axis indicates the correlation rank, and the vertical axis indicates VF(i). The FPO analysis unit 132 obtains the maximum value of VF(i) (final VF) for one element (i) (element of interest). In the example of Figure 6B, when the correlation rank is "5", there are three lines to elements within the module and two lines to elements outside the module, the degree is "5 = 2 + 3", the edge is 3, the number of elements in the module (n) is 5, and VF(i) is calculated from equation (2) above as "VF(i) = (2 × edge(i)) / (degree(i) + n - 1) = (2 × 3) / (5 + 5 - 1) = 0.67", which indicates the maximum value of VF(i) (final VF).

[0058] In addition, in calculating the final VF, the FPO analysis unit 132 may determine whether VF(i) does not exceed the maximum value even if the correlation order is increased. If the FPO analysis unit 132 determines that VF(i) does not exceed the maximum value even if the correlation order is increased, the FPO analysis unit 132 may end the process of calculating VF(i).

[0059] FIG. 6C is a virtual image diagram for explaining an example of a method for calculating the final VF according to this embodiment. For one FPO module, the FPO analysis unit 132 calculates a final VF for each element (i) in the FPO module. Specifically, the FPO analysis unit 132 sets each element (i) in the FPO module as an element of interest and calculates VF(i) for that element of interest. FIG. 6C shows the calculation result of VF(i) for one element (i) (element of interest). In FIG. 6C, the horizontal axis represents the correlation rank and the vertical axis represents VF(i). When the correlation rank is 8, the FPO analysis unit 132 determines that VF(i) does not exceed the maximum value and ends the process of calculating VF(i). The reason why the calculation of VF(i) can be ended is described below.

[0060]

number

[0061] In the above formula (3), VF is the element F score. n is the number of elements in the target module. edge is the number of edges in the target module. degree is the number of degrees of the elements. From the above formula (3), even if all elements in a module are connected to elements in other modules, edge = n-1, so we can say edge ≦ n-1. When calculating the final VF, the number of elements n in the module is a fixed value, so the upper limit of the achievable final VF can be found using only the number of degrees.

[0062]

number

[0063] In the above formula (4), VF is the element F score. a is the maximum VF value at the correlation rank a. n is the number of elements in the target module. degree is the number of degrees of the elements. In the example of FIG. 6C, n=5, and at the correlation rank a=5, VF a = 0.67, so VF aTo exceed this, d≦8 must be satisfied according to equation (4) above. Here, the value of degree depends on the correlation order; for a correlation order of "9," the degree value is 9. Therefore, after calculating the case of a correlation order of "8," calculations for correlation orders of "9" or higher are not necessary. In the example of FIG. 6C, for a correlation order of "6," there are three connections to elements within the module and three connections to elements outside the module, so the number of degrees (degree) is "6=3+3," the number of edges (edge) is 3, and the number of elements in the module (n) is 5. Therefore, according to equation (2) above, VF(i) is calculated as "VF(i)=(2×edge(i)) / (degree(i)+n-1)=(2×3) / (6+5-1)=0.60." For a correlation ranking of 7, there are three connections to elements within the module and four connections to elements outside the module, so the degree is 7 = 4 + 3, the number of edges is 3, and the number of elements in the module (n) is 5. Therefore, VF(i) is calculated from equation (2) above as follows: VF(i) = (2 × edge(i)) / (degree(i) + n-1) = (2 × 3) / (7 + 5-1) = 0.54. For a correlation ranking of 8, there are three connections to elements within the module and five connections to elements outside the module, so the degree is 8 = 5 + 3, the number of edges is 3, and the number of elements in the module (n) is 5. Therefore, VF(i) is calculated from equation (2) above as follows: VF(i) = (2 × edge(i)) / (degree(i) + n-1) = (2 × 3) / (8 + 5-1) = 0.50. In this way, as determined using formula (4), it can be confirmed that VF(i) does not exceed 0.67 (for correlation rank "5") when the element is an external element, even if the elements are added in descending order of correlation rank. By configuring in this way, it is possible to reduce the amount of calculation and shorten the processing time compared to when VF(i) is calculated for all correlation ranks.

[0064] The FPO analysis unit 132 calculates an MF (referred to as a final MF) for one FPO module based on the final VF of each element (i) in the FPO module. Specifically, the FPO analysis unit 132 acquires edge(i) and degree(i) when the final VF is obtained for each element (i) in the FPO module, and calculates the MF (final MF) according to the above formula (1) using the acquired edge(i) and degree(i) of each element (i).

[0065] The method for calculating the final VF and final MF may be any method that calculates VF(i) and MF without depending on the seed element in the module for which the final VF and final MF are to be calculated. For example, the VF(i) in the module where an element P value (described later) is the smallest among the other elements in the module may be used as the final VF.

[0066] (Step S6-2) The P value calculation unit 137 calculates a module P value for the FPO module of each seed element by the following formula.

[0067]

number

[0068] N is the total number of elements in the multivariate data that is the input data. n is the number of elements in the module. edge avg is the integer closest to the average value of edge(i) for each element(i) used in calculating the final MF. avg is the integer closest to the average value of the degree(i) of each element(i) used in calculating the final MF. q and r are positive integers, and r≦q. In addition, edge avg Instead of the number of edges in the module that connect elements with a correlation coefficient equal to or greater than the correlation threshold when the maximum value of the MF of the FPO module is obtained, the number of edges in the module that connect elements with a correlation coefficient equal to or greater than the correlation threshold when the maximum value of the MF of the FPO module is obtained is used. avg Instead of this, the degree at which the maximum value of MF of the FPO module is obtained may be used.

[0069] The module P value calculated for the FPO module of each seed element is stored as the module P value after FPO analysis in the evaluation value information 126. The module P value has the same meaning as probability used in statistical processing, and represents the probability that an event rarer than the observed event will occur. The module P value can be used to verify whether the formed module is correct. The module P value calculated by the above formula (5) is found by multiplying the probability of giving the same MD and the probability of giving the same MS when the module is randomly constructed. The module P-value after FPO analysis allows the operator to evaluate the significance of the FPO module of each seed element, thereby adding credibility to the module.

[0070] (Step S7) The FPO module integration unit 134 executes module integration processing (modularization processing) for the FPO modules of each seed element. An FPO module is constructed separately for each seed element through the above-mentioned steps S3-S5. Therefore, FPO modules are constructed for the total number of elements, and overlapping elements occur between the FPO modules. This overlapping element is resolved by the module integration processing. The module integration processing will be described below.

[0071] The FPO module integration unit 134 reconstructs modules for each VF threshold within a predetermined range of VF thresholds (from the minimum to the maximum VF thresholds). The VF thresholds are set in advance at fixed intervals from the minimum to the maximum VF thresholds. The module reconstructing method involves deleting all existing edges for all FPO modules and reconnecting elements with final VFs equal to or greater than the VF threshold with new edges. At this time, if there are overlapping elements between FPO modules, the FPO modules are also integrated via the overlapping elements. This allows the integration results of the FPO modules to be obtained for each VF threshold. The FPO module integration unit 134 stores the integration results of the FPO modules for each VF threshold in the module integration information 124. The VF threshold is not limited to being set at a constant interval from the minimum to the maximum value of the VF threshold. For example, all possible values ​​of MF and VF may be listed as the VF threshold, and all MF and VF may be integrated. In this case, the VF threshold does not need to be set at a constant interval.

[0072] The FPO module integration unit 134 may perform the integration process on only some of the FPO modules rather than on all of the FPO modules constructed in steps S3-S5. For example, it may exclude FPO modules that do not satisfy predetermined conditions based on the final VF, final MF, and module P value of each FPO module stored in the evaluation value information 126. For example, when the VF threshold is set from the minimum to maximum VF thresholds, it may exclude FPO modules whose final MF is less than the VF threshold and FPO modules whose seed element final VF is less than the VF threshold from the integration process. In this way, by excluding FPO modules with low quality or significance from the integration process based on the evaluation value, it is possible to improve the quality or significance of the modules being integrated.

[0073] FIG. 7 is a hypothetical conceptual diagram illustrating an example of the integration process according to this embodiment. FIG. 7A is a conceptual diagram illustrating the modularization process, in which the final VF is used to remove edges between elements and integrate modules. As an example, FIG. 7A shows three FPO modules M7-1, M7-2, and M7-3, with elements indicated by circles, element identifiers "c001"-"c011," and the final MF and final VF of each FPO module calculated in step S6-1. In this example, element "c002" overlaps between FPO modules M7-1 and M7-2, and element "c008" overlaps between FPO modules M7-2 and M7-3. When the VF threshold is set to 0.5, a module M7-4 including all elements is constructed by the integration process. When the VF threshold is set to 0.75, the final VFs of element "c002" in module M7-2 and element "c011" in module M7-3 are below the VF threshold, so they are not connected by an edge and are excluded by the integration process. As a result, modules M7-5 and M7-6 are formed by the integration process. When the VF threshold is set to 0.85, modules M7-7, M7-8, and M7-9 are formed when the integration process is performed in the same way. Here, modules M7-8 and M7-9 have only two and one element, respectively, so they are evaluated low and do not show significant connections between multiple elements, which hinders the operator's analysis of the integration results of the FPO modules.

[0074] On the other hand, Figure 7B is an illustration of the modularization process, in which components for each module are excluded using the final MF and the final VF of the seed element, and edges between elements are excluded using the final VF, thereby integrating modules. Figure 7B shows the integration process when the VF threshold is set from the minimum to maximum VF threshold, and FPO modules whose final MF is less than the VF threshold and whose seed element's final VF is less than the VF threshold are excluded from the integration process. When the VF threshold is set to 0.85, modules M7-2 and M7-3 whose final MF is less than 0.85 are excluded from the integration process, and only module M7-7 is constructed when the integration process is performed. In this way, the integration process can prevent the construction of modules with low evaluation values, thereby improving the quality or significance of the modules that are constructed.

[0075] The FPO module integration unit 134 stores the integration result (FPO module group) of the FPO modules with the VF threshold that has the maximum number of FPO modules that fit within a predetermined range of module size in the final module integration information 125 as the final FPO module integration result.

[0076] In the past, the final integration result of the FPO module alone did not provide an overall picture of the positional relationships of the elements, making it difficult to analyze the internal structure of the FPO module. However, according to this embodiment, the integration result of the FPO module for each VF threshold is stored in the module integration information 124, so that the operator can check how the integration result of the FPO module changes as the VF threshold changes, based on the integration result of the FPO module for each VF threshold. This allows the operator to grasp the overall picture of the positional relationships of the elements in the final integration result of the FPO module, which contributes to the analysis of the internal structure of the FPO module.

[0077] 8 and 9 are diagrams showing an example of the final FPO module integration results according to this embodiment. FIGS. 8 and 9 show the results of FPO analysis using a dataset showing the friendships of 34 members of a karate club at a university in the United States, with Pearson's product-moment correlation coefficient, a MIN value of 5, and a MAX value of 34. Components are shown as circles, and the patterns inside the circles distinguish each member. The numbers are the identification numbers of the 34 members of the karate club. The solid lines show the relationships between connections obtained by the modularization process. FIG. 8 shows the results of network depiction for each of the obtained modules. FIG. 9A is a diagram showing an example of the FPO module integration results for each VF threshold corresponding to the final FPO module integration results. In the example of FIG. 9A, a graph is shown showing how the FPO module integration results (modularization of each element (vertical axis)) change depending on the VF threshold (horizontal axis). For example, when the VF threshold (VF th ) is 0.81, the relationship between the modules M8-1, M8-2, and M8-3 for the three modules M8-1, M8-2, and M8-3 shown in FIG. 8 can be understood from FIG. 9A.

[0078] The output data generation unit 138 may use the module integration information 124 for each VF threshold to generate data (e.g., a graph) showing the correspondence between the VF threshold (horizontal axis) and the integration result of the FPO modules (modularization of each element (vertical axis)) as shown in FIG. 9. This graph data can be displayed on a display screen or printed out. This allows the operator to easily analyze the final integration result of the FPO modules.

[0079] The FPO module integration unit 134 may further calculate a distance index d indicating the distance of the final FPO module after module integration based on the integration results of the FPO modules for each VF threshold stored in the module integration information 124. The distance index is a positive real number and is calculated for all pairs of FPO modules. The lower the value of the distance index d, the closer the distance between the two modules is determined to be. Calculating the distance index d makes it possible to easily and quantitatively evaluate the overall positional relationship between the FPO modules.

[0080] 9B is a diagram for explaining an example of a method for calculating the distance index d of the final FPO module according to this embodiment. FIG. 9B shows one method for calculating the distance index d based on the integration result of the final FPO module exemplified in FIG. 9A. In FIG. 9B, the VF shown in FIG. 9A th For the three modules M8-1, M8-2, and M8-3 separated at =0.81, we consider virtual modules M9-1, M9-2, and M9-3, which include additional information about the module structure when the VF threshold is raised. Point G in the graph is defined as the center point of the separated virtual modules M9-1, M9-2, and M9-3. Regarding the distance between the obtained FPO modules, by calculating the distance to the center point of the virtual module using the difference in VF threshold and the number of components, we can more accurately grasp the overall picture of the positional relationships of the elements in the integration results of the FPO modules.

[0081] For example, when calculating the distance index between modules M9-1 and M9-2, the processing is performed according to the following steps (1) to (3). (1) The minimum VF threshold at which the components of modules M8-1 and M8-2 are separated by the module integration process, VF th,min In this example, VF th,min =0.76. (2) The components of virtual modules M9-1 and M9-2 set the VF threshold to VF th When set to above, the average VF of the minimum VF thresholds excluded by the module integration process th, minavgFor example, in the case of the virtual module M9-1, the elements "23", "21", "19", "16", and "15" are excluded if the VF threshold is 0.95 or more, the elements "33", "31", "30", and "24" are excluded if the VF threshold is 0.92 or more, the elements "34", "29", "28", "27", and "10" are excluded if the VF threshold is 0.91 or more, the element "9" is excluded if the VF threshold is 0.90 or more, and the element "32" is excluded if the VF threshold is 0.86 or more. Therefore, the average value of the minimum VF threshold at which 16 components are excluded is calculated as VF th, minavg (M8-1). Specifically, VF th, minavg This can be calculated as (M8-1) = (0.95 x 5 + 0.92 x 4 + 0.91 x 5 + 0.90 x 1 + 0.86 x 1) / 16 = 0.9212. Similarly, for virtual module M9-2, elements “8”, “4”, “22”, “2”, and “18” are excluded if the VF threshold is 0.90 or higher, elements “1”, “13”, and “14” are excluded if the VF threshold is 0.88 or higher, element “20” is excluded if the VF threshold is 0.87 or higher, and element “3” is excluded if the VF threshold is 0.86 or higher. Therefore, the average VF of the minimum VF threshold is 10, and 10 components are excluded. th, minavg (M8-2). Specifically, VF th, minavg This can be calculated as (M8-2)=(0.9×5+0.88×3+0.87×1+0.86×1) / 10=0.8870. (3) Calculate the distance index d(M8-1, M8-2) using the following equation (6).

[0082]

number

[0083] In the above equation (6), d indicates the distance index between modules. th indicates the VF threshold. th,min indicates the minimum VF threshold at which the components of modules M8-1 and M8-2 are separated by the module merging process. th, minavgindicates the average of the minimum VF thresholds at which components of both modules are excluded. In this example, d(M8-1, M8-2) = 2(0.81-0.76) + (0.8870-0.81) / 2 + (0.9212-0.81) / 2 = 0.1000 + 0.0385 + 0.0556 = 0.1941.

[0084] When the same calculation is performed for the virtual modules M9-2 and M9-3, d(M8-2, M8-3) = 0.0585 and d(M8-1, M8-3) = 0.1756 are calculated. In the case of d(M8-2, M8-3), for the virtual module M9-3, the elements "7", "6", "5", "12", and "11" are excluded when the VF threshold is 0.85 or more. Therefore, the average value of the VF threshold at which the five components are excluded, VF th, minavg (M8-3). Specifically, VF th, minavg (M8-3) = 0.85 × 5 / 5 = 0.8500. The minimum VF threshold, VF th,min In this example, VF th,min = 0.81. In this example, we can calculate d(M8-2, M8-3) = 2(0.81 - 0.81) + (0.8870 - 0.81) / 2 + (0.8500 - 0.81) / 2 = 0.0385 + 0.0200 = 0.0585. In addition, in the case of d(M8-1, M8-3), the minimum VF threshold at which the components of modules M8-1 and M8-3 are divided by the module integration process is VF th,min In this example, VF th,min = 0.76. In this example, we can calculate d(M8-1, M8-3) = 2(0.81 - 0.76) + (0.9212 - 0.81) / 2 + (0.8870 - 0.81) / 2 = 0.1000 + 0.0556 + 0.0200 = 0.1756. From the above, it can be determined that M8-2 is closer to M8-1 than M8-3. In this way, it is possible to more accurately grasp the overall picture of the positional relationship of elements in the final FPO module integration result.

[0085] 9B shows distance indices calculated based on the final integration results of the FPO modules shown in FIG. 9A. For module M8-1, the distance 0.1056 indicates the average distance of each element from the minimum VF threshold at which module M8-2 is divided to the minimum VF threshold at which the elements constituting module M8-1 are excluded. For module M8-1, the distance 0.0556 indicates the average distance from the minimum VF threshold at which module M8-3 is divided to the minimum VF threshold at which the elements constituting module M8-1 are excluded. For module M8-2, the distance 0.0885 indicates the average distance of each element from the minimum VF threshold at which module M8-1 is divided to the minimum VF threshold at which the elements constituting module M8-2 are excluded. For module M8-2, the distance 0.0385 indicates the average distance of each element from the minimum VF threshold at which module M8-3 is divided to the minimum VF threshold at which the elements constituting module M8-2 are excluded. In module M8-3, the distance 0.0700 indicates the average distance from the minimum VF threshold at which module M8-1 is divided to the minimum VF threshold at which the elements constituting module M8-3 are excluded. In module M8-3, the distance 0.0200 indicates the average distance of each element from the minimum VF threshold at which module M8-2 is divided to the minimum VF threshold at which the elements constituting module M8-3 are excluded.

[0086] (Step S8-1) The FPO module integration unit 134 calculates a final VF and a final MF for each of the final FPO modules after the module integration, similar to the FPO analysis unit 132 in step S6-1 above. The final VF and the final MF calculated for each of the final FPO modules after the module integration are stored in the evaluation value information 126.

[0087] (Step S8-2) The P value calculation unit 137 calculates a module P value for each final FPO module after the module integration using the above formula (5). The module P value calculated for each final FPO module after the module integration is stored in the evaluation value information 126 as the module P value after the module integration. The module P-value after module integration allows the operator to evaluate the significance of each final FPO module after module integration, thereby adding to the reliability of the module.

[0088] (Step S9) The FNI analysis unit 135 performs a process (FNI analysis process) to incorporate (In) false negative elements for some or all of the final FPO modules (final FPO modules) after module integration.

[0089] For example, the user may be allowed to specify multiple final FPO modules to be the target of the FNI analysis process. Alternatively, conditions for the final FPO modules to be the target of the FNI analysis process (FNI analysis process target conditions) may be set in advance in the information processing device 1, and the FNI analysis unit 135 may exclude final FPO modules that do not satisfy the FNI analysis process target conditions and determine the remaining final FPO modules as the target of the FNI analysis process. The FNI analysis process target condition is, for example, a module having a certain level of significance (module P value) or higher. The FNI analysis process is described below.

[0090] FIG. 10 is a virtual image diagram for explaining the FNI analysis processing according to this embodiment. In FIG. 10, elements are represented by circles. The elements include a focus element, a candidate element, elements in the target module, and elements outside the target module. Elements in the target module are elements that will be included in the final FPO module (final FPO module) after module integration, and elements outside the target module are elements that will not be included in the final FPO module (final FPO module) after module integration. The numbers shown in the circles indicate the correlation order with respect to the candidate element. Elements are connected by lines. Solid lines indicate the relationship between elements in the module, including the focus element. Dotted lines indicate the relationship between elements outside the module, including the candidate element. The FNI analysis unit 135 repeats the FNI analysis process for the element of interest while sequentially setting all elements included in all final FPO modules that are the target of the FNI analysis process as the element of interest. Therefore, the FNI analysis process is executed as many times as the number of elements included in all final FPO modules that are the target of the FNI analysis process. Here, with reference to FIG. 10, an explanation will be given using one element of interest shown in FIG. 10 as an example.

[0091] The FNI analysis unit 135 sets candidate elements to be incorporated into the final FPO module (target module) to which the element of interest belongs. For example, the candidate elements to be incorporated into the target module may be all elements that do not belong to the target module (i.e., all elements outside the target module). For example, the candidate elements to be incorporated into the target module may be limited to those elements outside the target module that have a correlation ranking with the element of interest that is at a predetermined high level. For example, the candidate elements to be incorporated into the target module may be limited to those elements outside the target module that have a correlation coefficient with the element of interest that is equal to or greater than a predetermined correlation threshold. For example, the candidate elements to be incorporated into the target module may be limited to those elements outside the target module that have a correlation ranking with the element of interest that is at a predetermined high level and whose correlation coefficient with the element of interest is equal to or greater than a predetermined correlation threshold. For example, the candidate elements to be incorporated into the target module may be limited to those elements that do not belong to any final FPO module. The FNI analysis unit 135 performs a false negative determination for the target module for each candidate element.

[0092] Specifically, the FNI analysis unit 135 references the correlation coefficient list 122 for the candidate element i in order from the top (ranking from highest correlation (correlation rank)), and for each rank, calculates VS(i) for the target module of the candidate element i, "VS(i) = (number of edges in the target module of the candidate element i, edge(i)) / (number of degrees of the candidate element i, degree(i))" of the above formula (2). The FNI analysis unit 135 records information such as correlation rank for VS(i) that exceeds a predetermined VS threshold in the storage unit 12 among all the calculated VS(i). The VS threshold can be set arbitrarily by, for example, an operator. If the calculated VS(i) exceeds the predetermined VS threshold, the FNI analysis unit 135 determines that the candidate element is a false negative element for the element of interest, and incorporates the candidate element into the target module.

[0093] (Step S10) The P-value calculation unit 137 calculates the element P-value for the target module in the correlation order corresponding to VS(i) that exceeds the VS threshold value, using the following formula.

[0094]

number

[0095] In the above equation (7), N is the total number of elements in the multivariate data that is the input data. n is the number of elements in the target module. edge is the number of edges edge(i) in the target module for candidate element i. degree is the number of degrees degree(i) of candidate element i. q and r are positive integers, and r≦q.

[0096] The element P value may be calculated using the recursive formula of equation (8) while calculating VS(i) at each correlation rank in step S9.

[0097]

number

[0098] In the above equation (8), N is the total number of elements in the multivariate data that is the input data. n is the number of elements in the module. edge is the number of edges edge(i) in the target module of candidate element i. degree is the number of degrees degree(i) of candidate element i. By calculating the element P value using the recursive formula of equation (8) while calculating VS(i) in step S9, it is possible to omit the calculation of the number of combinations, thereby improving the processing efficiency.

[0099] 11 shows the time required to perform FNI analysis (including steps S9 to S13-2 (described below)) after completing FPO analysis using Arabidopsis microarray data (a data set of 27,127 elements x 9,442 samples) with Pearson's product-moment correlation coefficient, MIN value of 2, and MAX value of 250. As in the present invention, steps S9 and S10 were executed simultaneously, and a recursive formula was used to calculate the element P value, significantly reducing the analysis time to 1 / 92.

[0100] The P value calculation unit 137 stores information about the correlation rank with the smallest element P value among the correlation ranks corresponding to VS(i) that exceed the VS threshold in the FNI analysis evaluation value information 126. The information about the correlation rank with the smallest element P value is information about the correlation rank that can be determined to be most significant.

[0101] (Step S11) If the FNI analysis process has been completed for all elements of all final FPO modules that are the target of the FNI analysis process (YES in step S11), proceed to step S12. On the other hand, if there are elements for which the FNI analysis process has not yet been performed (NO in step S11), return to step S9 and perform the FNI analysis process on the elements for which the FNI analysis process has not yet been performed, as the element of interest.

[0102] (Step S12) The FNI assignment processing unit 136 executes an assignment process to eliminate overlapping candidate elements among multiple final FPO modules generated by the FNI analysis process. In the assignment process, the following assignment criteria are used.

[0103] Assignment criteria: For multiple target modules with overlapping candidate elements i, the following priorities are used: VS(i)> "The module P value of the above formula (5) when candidate element i is incorporated into the target module"> Number of elements in the module VS(i) has the highest priority, followed by the module P value, and then the number of elements in the module.

[0104] First, candidate element i is left only in the target module with the largest VS(i), and candidate element i is removed from the other target modules. On the other hand, if VS(i) is equivalent, the candidate element i is retained only in the target module with the smallest "module P value of equation (5) above when the candidate element i is incorporated into the target module," and the candidate element i is deleted from the other target modules. If VS(i) is equivalent and the "module P value of equation (5) above when candidate element i is incorporated into the target module" is equivalent between the target modules, candidate element i is retained only in the target module with the largest number of elements, and candidate element i is deleted from the other target modules.

[0105] The assignment process is not limited to the process based on the above-mentioned assignment criteria, as long as it can eliminate overlapping elements between multiple modules that occur as a result of the FNI analysis process. For example, modules with overlapping elements may be integrated by a module integration process (modularization process).

[0106] The FNI assignment processing unit 136 stores information on the final FPO module after the assignment process in the final module information 127. The final module information 127 may be a file for each module, a file that compiles all modules, or both a file for each module and a file that compiles all modules. Note that the information on the final FPO module before the assignment process may be recorded in the storage unit 12.

[0107] According to this embodiment, when duplicate candidate elements occur in multiple final FPO modules as a result of performing FNI analysis processing on multiple final FPO modules, it is possible to eliminate the duplication of the candidate elements. This saves the operator the trouble of individually specifying the final FPO modules on which to perform the FNI analysis processing, and it is possible to perform the FNI analysis processing on multiple final FPO modules at once, and to obtain FNI analysis processing results without duplicate elements.

[0108] (Step S13-1) The FNI assignment processing unit 136 calculates a final VF and a final MF for each final FPO module after the assignment process, similar to the FPO analysis unit 132 in step S6-1 above. The final VF and final MF calculated for each final FPO module after the assignment process are stored in the evaluation value information 126.

[0109] (Step S13-2) The P value calculation unit 137 calculates a module P value for each final FPO module after the assignment process using the above formula (5). The module P value calculated for each final FPO module after the assignment process is stored in the evaluation value information 126 as the module P value after the assignment process. The module P-value after the assignment process allows the operator to evaluate the significance of each final FPO module after the assignment process, thereby adding credibility to the module.

[0110] (Step S14) The output data generation unit 138 generates various output data. The output data is data that can be displayed on a display screen or printed. For example, the output data generation unit 138 generates drawing data that represents the configuration of each final FPO module after the assignment process. The configuration of each final FPO module after the assignment process is visualized by drawing it using this drawing data.

[0111] For example, the output data generation unit 138 may generate graph data (such as the graph data illustrated in FIG. 5A) showing the MF corresponding to each rank for each seed element obtained by the FPO analysis process. For example, the output data generation unit 138 may generate graph data (such as the graph data illustrated in FIG. 9) showing the integration results for each VF threshold obtained by the FPO module integration process.

[0112] The output data generated by the output data generation unit 138 may be output by the input / output unit 11 to a device external to the information processing device 1, or may be stored in the storage unit 12. The output data stored in the storage unit 12 can be output by the input / output unit 11 to an external device.

[0113] According to this embodiment, it is possible to obtain an effect of improving the accuracy of correlation network analysis of multivariate data such as comprehensive molecular information obtained by omics analysis or big data.

[0114] Although an embodiment of the present invention has been described in detail above with reference to the drawings, the specific configuration is not limited to this embodiment, and design changes and the like are also included within the scope that does not deviate from the gist of the present invention.

[0115] In addition, a computer program for realizing the functions of each of the above-described devices may be recorded on a computer-readable recording medium, and the program recorded on the recording medium may be read and executed by a computer system. Note that the "computer system" here may also include hardware such as an OS and peripheral devices. Furthermore, if a WWW system is used, the "computer system" also includes the homepage provision environment (or display environment). In addition, "computer-readable recording medium" refers to writable non-volatile memory such as a flexible disk, optical magnetic disk, ROM, or flash memory, portable media such as a DVD (Digital Versatile Disc), or a storage device such as a hard disk built into a computer system.

[0116] Furthermore, the term "computer-readable recording medium" also includes a device that stores a program for a certain period of time, such as volatile memory (e.g., DRAM) within a computer system that serves as a server or client when a program is transmitted via a network such as the Internet or a communication line such as a telephone line. The program may be transmitted from a computer system storing the program in a storage device or the like to another computer system via a transmission medium or by transmission waves in the transmission medium. Here, the "transmission medium" that transmits the program refers to a medium that has the function of transmitting information, such as a network (communication network) such as the Internet or a communication line (communication line) such as a telephone line. The program may also be a program for implementing some of the functions described above, or may be a so-called differential file (differential program) that can implement the functions described above in combination with a program already stored in the computer system. [Explanation of symbols]

[0117] 1...information processing device, 11...input / output unit, 12...storage unit, 121...correlation network analysis program, 122...correlation coefficient list, 123...module information, 124...module integration information, 125...final module integration information, 126...evaluation value information, 127...final module information, 128...FPO analyzed module list, 13...control unit, 131...correlation coefficient calculation unit, 132...FPO analysis unit, 133...FPO module selection unit, 134...FPO module integration unit, 135...FNI analysis unit, 136...FNI assignment processing unit, 137...P value calculation unit, 138...output data generation unit

Claims

1. On the computer, When constructing a module of seed elements from non-seed elements having a high correlation with a seed element among elements contained in multivariate data that are the subject of correlation network analysis, an FPO analysis process is performed for each rank, removing false positive non-seed elements from the non-seed elements having a high correlation with the seed element, while increasing the rank from a lower limit value to an upper limit value, which is the number of elements other than the seed element to be incorporated into the module of the seed element; storing configuration information of the module on which the FPO analysis process has been executed in a storage unit; If a module identical to the module configuration information stored in the storage unit is constructed, the FPO analysis process is stopped and the search for the module showing the maximum module F score of the rank is terminated early. program.

2. The computer, The correlation coefficient list, the module information for each rank by element, the module F score and module integration information for each element F score threshold, the final module integration information, the evaluation value information for each target step, the final module information, and the FPO analyzed module list are stored in the storage unit, Calculating a hash value for each piece of information stored in the storage unit and adding it to the information; Each piece of information is identified by the hash value. The program according to claim 1.

3. The computer, Regarding the module F score at each rank from the lower limit to the upper limit of the rank, the module with the rank that indicates the maximum value of the period during which the best value of the module F score is not updated even if the rank is increased is selected as the optimal module of the FPO analysis result; If a predetermined condition is satisfied, the FPO analysis process and module selection are terminated. The program according to claim 1.

4. The conditions for terminating the FPO analysis process and module selection are a comprehensive criterion based on the number of ranks for which module F scores have not been calculated, the maximum period for which the best module F score has not been updated, and the period for which the module F score has not been updated at the time of the current rank. The program according to claim 3.

5. The computer, Calculating an element F-score, which is an index for evaluating the components of the module, for each of the multiple elements constituting the module; Calculating a module F score, which is an index for evaluating the entire module, based on the maximum value of the element F scores; At the maximum value of the element F score, it is determined whether the element F score does not exceed the maximum value even when the correlation rank is increased; If it is determined that the maximum value is not exceeded even if the correlation rank is increased, the process of calculating the element F score is terminated. The program according to claim 1.

6. The criterion for determining whether the maximum value of the element F score is exceeded is a comprehensive criterion based on the number of elements in the module and elements inside and outside the module. The program according to claim 5.

7. The computer, When reconstructing modules for each seed element configured using each element of the multivariate data as a seed element for each threshold of the element F score, elements having an element F score equal to or greater than the threshold of the element F score are associated with each other and modules are integrated via overlapping elements; Calculating an element F-score, which is an index for evaluating the components of a module, for each of the multiple elements that make up the integrated module; Calculating a module F score, which is an index for evaluating the entire module, based on the maximum value of the element F scores; At the maximum value of the element F score, it is determined whether the element F score does not exceed the maximum value even when the correlation rank is increased; If it is determined that the maximum value is not exceeded even if the correlation rank is increased, the process of calculating the element F score is terminated. The program according to claim 1.

8. The criterion for determining whether the maximum value of the element F score is exceeded is a comprehensive criterion based on the number of elements in the module and elements inside and outside the module. The program according to claim 7.

9. On the computer, Execute an FNI analysis process that incorporates false negative elements into a plurality of modules composed of elements contained in the multivariate data that is the subject of correlation network analysis; Simultaneously with the FNI analysis process, the element P value is calculated using the following formula while calculating the element specificity rate at each correlation order of the candidate element, which is an element outside the module, for the element of interest. [Equation 1] N is the total number of elements in the multivariate data, n is the number of elements in the target module, edge is the number of edges of the candidate element within the target module, degree is the degree of the candidate element, program.

10. The computer, The correlation coefficient list, the module information for each rank by element, the module F score and module integration information for each element F score threshold, the final module integration information, the evaluation value information for each target step, the final module information, and the FPO analyzed module list are stored in a storage unit, Calculating a hash value for each piece of information stored in the storage unit and adding it to the information; Each piece of information is identified by the hash value. The program according to claim 9.

11. The computer, Execute an assignment process to eliminate overlapping elements in multiple modules, Calculating an element F-score, which is an index for evaluating the components of the module, for each of the multiple elements that make up the module for which the assignment process has been executed; Calculating a module F score, which is an index for evaluating the entire module, based on the maximum value of the element F scores; At the maximum value of the element F score, it is determined whether the element F score does not exceed the maximum value even if the correlation rank is increased, and when it is determined that the element F score does not exceed the maximum value even if the correlation rank is increased, the process of calculating the element F score is terminated. The program according to claim 9.

12. The criterion for determining whether the maximum value of the element F score is exceeded is a comprehensive criterion based on the number of elements in the module and elements inside and outside the module. The program according to claim 11.

Images