Pathway generation device, pathway generation method, and pathway generation program

Abstract

When a pathway is generated on the basis of property information indicating properties of molecules that act on a disease or a symptom, interactome information indicating connection relationships among molecules, and similarity information indicating degrees of similarity among molecules, a pathway is generated by using a minimum flow algorithm so as to satisfy the following at least two conditions: (1) the pathway is to be generated by connecting only molecules with which the flow rate value becomes greater than a minimum threshold and less than a maximum threshold; and (2) the pathway is to be generated by connecting only molecules with which the flow rate value becomes less than or equal to a threshold relating to the degree of similarity. Accordingly, when a pathway is regenerated by knocking out specific molecules in a pathway once generated, changes occur in other pathways as well as in pathways on which the knocked-out molecules are present, which makes it possible to generate an appropriate pathway reflecting the knocked-out state.

Description

Pathway generation device, pathway generation method, and pathway generation program

[0001] The present invention relates to a pathway generation device, a pathway generation method, and a pathway generation program, and is particularly suitable for use in a technique for generating pathways that represent intermolecular interactions as a route diagram.

[0002] Pathways (also called molecular networks) are known as diagrams that represent molecular interactions. Pathways are represented by symbols such as circles and squares, with arrows connecting the symbols to indicate molecular interactions. Visualizing molecular interactions in this way makes it easier to understand biological phenomena, for example by identifying which pathways contain genes with altered expression levels. Pathways are widely used in fields such as disease treatment and drug discovery.

[0003] A technology for generating pathways by identifying the intermolecular connectivity using a minimum flow algorithm is known (see, for example, Patent Document 1). The pathway generation device described in Patent Document 1 generates pathways by placing causative molecules upstream and responsive molecules downstream, reflecting the intermolecular connectivity indicated in a known knowledge database, and by making it easier for molecules with similarity values ​​between disease feature vectors and molecular feature vectors to be connected. In this case, the minimum flow algorithm is used to identify sequential connectivity between three or more molecules not indicated in the known knowledge database.

[0004] Also known is a technology that generates an intermolecular network of any range in response to a user's instructions (see, for example, Patent Document 2). Patent Document 2 discloses, as an example, a method in which one or more molecules are specified in an intermolecular network that has already been generated, a connect search is performed again, and the intermolecular network generated there (which becomes a partial network of the original intermolecular network) is marked and displayed. This method is considered useful for predicting the range of an intermolecular network that will be affected, for example, when the expression of a certain protein is suppressed by a method such as knockout or RNA interference, or when the function of a certain protein is modified by a transgenic method (also known as knock-in).

[0005] However, the technology described in Patent Document 2, as shown in Figure 12 of Patent Document 2, merely obtains and marks (highlights molecules belonging to the partial network) a partial network that connects a molecule designated by the user as a starting point (indicated by one * mark in Figure 12) and a molecule designated by the user as an end point (indicated by two # marks in Figure 12) in an intermolecular network that has been generated once, by a connect search. Therefore, for example, when a certain molecule is knocked out or knocked in, it is not possible to analyze what effect occurs on the partial network other than the partial network that connects the starting point and end point designated by the user.

[0006] Patent No. 6915818 Publication Special Publication No. 2008-515029

[0007] The present invention has been made to solve these problems, and aims to make it possible to analyze what effect occurs on the entire pathway (intermolecular network) when a specific molecule is knocked out in the pathway (intermolecular network) once it has been generated.

[0008] To solve the above-mentioned problems, the present invention generates a pathway that represents intermolecular interactions as a path diagram based on property information that represents the properties of molecules that act on a disease or symptom, connection information that represents the connection relationships between molecules, and similarity information that represents the similarity between molecules, so that causative molecules are located upstream and responsive molecules are located downstream, while also reflecting the connection relationships represented by the connection information and making it easier for molecules with high similarity represented by the similarity information to be connected. The pathway is generated using a minimum flow algorithm so as to satisfy at least two conditions: (1) generating a pathway by connecting only molecules whose flow values ​​are greater than a minimum threshold and less than a maximum threshold, and (2) generating a pathway by connecting only molecules whose flow values ​​are equal to or less than a threshold related to similarity.

[0009] According to the present invention configured as described above, not only are pathways passing through molecules with flow values ​​below a minimum threshold or above a maximum threshold eliminated, but pathways connecting molecules with low similarity and pathways passing through molecules with flow values ​​exceeding a threshold related to similarity are also eliminated. As a result, the entire pathway (intermolecular network) is generated by selecting intermolecular connections based on the magnitude of flow values ​​and similarity from among the intermolecular connections defined by the connection information representing the connection relationships between molecules.

[0010] Therefore, when a specific molecule is knocked out of a pathway that has already been generated and the pathway is regenerated, changes occur not only in the pathway in which the knocked-out molecule existed but also in other pathways according to changes in the relationship between the redistributed flow value and the threshold, and an appropriate pathway that reflects the knocked-out state is generated. As described above, according to the present invention, it is possible to analyze what effect occurs on the entire pathway when a specific molecule is knocked out of a pathway that has already been generated.

[0011] FIG. 1 is a block diagram showing an example of the functional configuration of a pathway generation device according to this embodiment. FIG. 2 is a block diagram showing an example of the functional configuration of a feature vector calculation device. FIG. 3 is a diagram showing an example of a molecular feature vector. FIG. 4 is a schematic diagram for explaining conditions (1) and (2) when executing a minimum flow algorithm. FIG. 5 is a diagram showing an example of a pathway generation method. FIG. 6 is a diagram showing an example of regenerating a pathway by knocking out one molecule included in a pathway that has already been generated. FIG. 7 is a block diagram showing another example of the functional configuration of a pathway generation device according to this embodiment.

[0012] An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing an example of the functional configuration of a pathway generation device 10 according to this embodiment. As shown in FIG. 1, the pathway generation device 10 of this embodiment includes a pathway generation unit 11 as a functional configuration. In addition, a connection information storage unit 12, a property information storage unit 13, and a similarity information storage unit 14 are connected to the pathway generation device 10 of this embodiment as storage media.

[0013] The pathway generation unit 11 can be configured using any of hardware, a DSP (Digital Signal Processor), and software. For example, the pathway generation unit 11 is realized by the operation of a pathway generation program stored in a storage medium such as RAM, ROM, a hard disk, or a semiconductor memory under the control of a microcomputer configured with a CPU, RAM, ROM, etc.

[0014] The connection information storage unit 12 stores connection information that represents the connection relationships between molecules. The connection information includes known information such as interactome information. The interactome information is known information regarding molecular interactions, such as, for example, when the expression level of a certain molecule (gene or protein) increases (or decreases), the expression level of another molecule also increases (or decreases) in conjunction with the increase (or decrease) of the expression level of a certain molecule.

[0015] Note that this interactome information only indicates which molecules are related to each other, and does not include information indicating the magnitude of the connection. Furthermore, interactome information is a collection of information indicating the connection between two molecules, and does not indicate the sequential connection between three or more molecules.

[0016] The property information storage unit 13 stores property information that indicates the properties of molecules that act on diseases or symptoms. The property of a molecule indicates whether the molecule acts as a causative or responsive molecule on a disease or symptom. Causative property is a property that the presence or mutation of the molecule may cause a disease or symptom. Responsive property is a property that the molecule may mutate due to the onset of a disease or symptom.

[0017] When generating a pathway related to a specific disease (hereinafter referred to as a disease pathway), it is possible to use known information recorded in various literature, databases, etc. as information on multiple molecules related to the disease and information on the properties (causality or responsiveness) of the actions of these molecules on the disease. It is also possible to estimate molecules related to a specific disease using a predetermined algorithm and use information on the properties of the actions of specific molecules on the disease estimated using the predetermined algorithm.

[0018] Any algorithm can be used as an algorithm for predicting disease-related molecules or molecular properties. For example, a new molecule related to a disease can be predicted by a prediction model trained by machine learning using known information about the association between a disease and a molecule. Furthermore, the properties of a new molecule can be predicted by a prediction model trained by machine learning using known information about the properties of molecules that act on a disease.

[0019] The same applies to generating a pathway related to a specific symptom (hereinafter referred to as a symptom pathway). That is, known information recorded in various literature, databases, etc. can be used as information on multiple molecules related to the symptom and information on the properties of the actions of those molecules on the symptom. It is also possible to estimate molecules related to a specific symptom using a predetermined algorithm and use information on the properties of the actions of the specific molecules on the symptom estimated using the predetermined algorithm.

[0020] The similarity information storage unit 14 stores similarity information representing the similarity between molecules. Here, the similarity information storage unit 14 stores information representing the similarity between each of a plurality of molecules stored as interactome information in the connection information storage unit 12. The similarity between molecules can be evaluated using various methods. For example, a method can be applied in which predetermined features are extracted from information about a plurality of molecules (e.g., descriptions of the molecules, structures or functions of the molecules, etc.) and the similarity of the features is evaluated. As an example, the similarity between pairs of molecules expressed by fingerprinting can be used.

[0021] As another example, molecular feature vectors can be calculated by analyzing multiple sentences (such as literature information) that describe information about molecules, and the similarity of the molecular feature vectors can be used. A molecular feature vector is data that represents the features (features that can identify molecules) possessed by molecules such as proteins and genes as a combination of multiple element values. As an example, a vector that represents the degree to which a molecule name contained as a word in multiple sentences contributes to each sentence is used as the molecular feature vector.

[0022] Molecular names as words tend to be used in sentences describing molecules, but not in sentences unrelated to molecules. Furthermore, among sentences describing molecules, sentences containing a certain molecular name as a word are likely to describe that molecule, while sentences describing other types of molecules are likely to not contain that molecular name. In other words, sentences containing molecular names as a word tend to differ depending on the type of molecule they are about. Therefore, a vector that represents the degree to which a molecular name contributes to each sentence can be used as a feature vector capable of identifying molecules.

[0023] Such a molecular feature vector can be calculated, for example, using the algorithm described in Japanese Patent No. 6915818. A method for calculating a molecular feature vector using the algorithm described in Japanese Patent No. 6915818 will be briefly described below. Figure 2 is a block diagram showing an example of the functional configuration of a feature vector calculation device 200.

[0024] In FIG. 2 , the word extraction unit 201 analyzes m sentences (m is any integer equal to or greater than 2) and extracts n words (n is any integer equal to or greater than 2) from the m sentences. The sentence analysis can be performed using, for example, a known morphological analysis. Note that the m sentences may contain the same word multiple times. In this case, the word extraction unit 201 does not extract the same word multiple times, but extracts only one word. In other words, the n words extracted by the word extraction unit 201 mean n types of words.

[0025] The vector calculation unit 202 calculates m sentence vectors and n word vectors from m sentences and n words. Here, the sentence vector calculation unit 202A converts each of the m sentences analyzed by the word extraction unit 201 into q-dimensional vectors according to a predetermined rule, thereby generating m sentence vectors d ｉ→ (i=1, 2, ..., m) (the symbol "→" indicates a vector). The word vector calculation unit 202B vectorizes each of the n words extracted by the word extraction unit 201 into q dimensions according to a predetermined rule, thereby calculating n word vectors w consisting of q axis components. ｊ → (j = 1, 2, ..., n) is calculated. ｉ → and word vector w ｊ →We will not explain the specific calculation method here.

[0026] The index value calculation unit 203 calculates the m text vectors d ｉ → and n word vectors w ｊ By taking the inner product of each of them, m sentences d ｉ and n words w ｊ Here, the index value calculation unit 203 calculates an index value that reflects the relationship between m sentence vectors d ｉ →each q axis component (d 11 ~d mq ) and n word vectors w ｊ →each q axis component (w 11 ~w nq ) as elements of a word matrix W, and calculates an index value matrix DW having m×n index values ​​as elements. ｔ is the transpose of the word matrix.

[0027]

[0028] Each element of the index value matrix DW calculated in this way can be said to represent the degree to which a word contributes to a sentence, and the degree to which a sentence contributes to a word. For example, the element dw in the first row and second column 12can be said to be a value that represents the degree to which word w2 contributes to sentence d1, and also a value that represents the degree to which sentence d1 contributes to word w2. As a result, each row of the index value matrix DW can be used to evaluate the similarity of sentences, and each column can be used to evaluate the similarity of words.

[0029] The feature vector identification unit 204 identifies, for each of a plurality of molecular names among the n words, a word index group consisting of m indexes for one molecular name as a molecular feature vector. That is, as shown in Fig. 3, the feature vector identification unit 204 identifies, as a molecular feature vector for each molecular name, a word index group for a word corresponding to a molecular name among n sets of word index groups (m indexes per column) constituting each column of the index matrix DW.

[0030] The similarity of molecular feature vectors can be evaluated in various ways. For example, a method can be applied in which a feature amount is extracted for each molecular feature vector using a predetermined function and the similarity of the feature amount is evaluated. Alternatively, the Euclidean distance or cosine similarity between molecular feature vectors may be used, or the edit distance may be used.

[0031] Returning to Figure 1, the pathway generation unit 11 generates a pathway that represents intermolecular interactions as a path diagram by an optimization process using a minimum flow algorithm, based on property information that represents the properties of molecules that act on a specific disease or symptom specified by, for example, a user, connection information (interactome information) that represents the connection relationships between molecules related to the specified disease or symptom, and similarity information that represents the similarity between molecules. The interactome information, property information, and similarity information are read out and used from the connection information storage unit 12, property information storage unit 13, and similarity information storage unit 14, respectively.

[0032] In this case, the pathway generation unit 11 generates a pathway by positioning causative molecules (hereinafter referred to as causative molecules) indicated by the property information upstream and responsive molecules (hereinafter referred to as responsive molecules) downstream, and by positioning other molecules (hereinafter referred to as connecting molecules) between the causative molecules and responsive molecules in a manner that reflects the connection relationships indicated by the interactome information, and by making it easier for molecules with high similarity indicated by the similarity information to connect to each other. Here, the pathway generation unit 11 sets one start point further upstream of all causative molecules and one end point further downstream of all responsive molecules, and applies the minimum flow algorithm to the section from the start point to the end point.

[0033] When a specific disease is designated by the user, the pathway generation unit 11 generates an intermolecular network representing, as a disease pathway, the intermolecular interactions of molecules related to the designated disease as a pathway diagram. When a specific symptom is designated by the user, the pathway generation unit 11 generates, as a symptom pathway, an intermolecular network representing, as a pathway diagram, the intermolecular interactions of molecules related to the designated symptom.

[0034] In this embodiment, the pathway generation unit 11 applies the minimum flow algorithm so as to satisfy at least the following two conditions: (1) A pathway is generated by connecting only molecules whose flow values ​​are greater than a minimum threshold and less than a maximum threshold, and (2) A pathway is generated by connecting only molecules whose flow values ​​are equal to or less than a threshold related to similarity.

[0035] Here, the flow value is an evaluation value calculated for each molecule when connecting each molecule using the minimum flow algorithm. This evaluation value is calculated to satisfy the following conditions: (A) The flow value of the starting point set upstream of the causative molecule is distributed sequentially to each causative molecule, connecting molecule, and responsive molecule, and the flow value aggregated at the end point set downstream of the responsive molecule is the same as the flow value of the starting point. The flow value of each molecule is smaller than the flow values ​​of the starting point and the end point. (B) The magnitude of the flow value of each molecule is determined taking into account the similarity between molecules. For example, the flow value is determined based on a predetermined function or calculation algorithm that uses similarity as a variable. This function or calculation algorithm is designed so that the greater the similarity, the larger the flow value. (C) The flow value is normalized to a value between 0 and 1. The magnitude of the similarity between molecules is also normalized to a value between 0 and 1. In this case, the flow value may be equal to the similarity between molecules.

[0036] Regarding condition (1), for example, the minimum threshold of the flow value can be set to "0." This means that pathways including molecules with a flow value of "0" are not adopted, and the flow values ​​of each molecule included in the pathway to be generated will not be "0." The maximum threshold of the flow value in condition (1) may be a value different from or the same as the threshold related to similarity in condition (2). If the values ​​are the same, this means that the maximum threshold of the flow value in condition (1) is determined by the threshold related to similarity (condition (2) is incorporated into condition (1)).

[0037] Figure 4 is a schematic diagram for explaining the above conditions (1) and (2). In the pathway shown in Figure 4, diamond-shaped nodes represent causal molecules, square-shaped nodes represent responsive molecules, and oval-shaped nodes represent connecting molecules (the same applies to other figures described later).

[0038] 4(a), two possible routes are possible based on the interactome information for the route from the causative molecule 41 to the responsive molecule 46. That is, a first route is possible from the causative molecule 41 to the connecting molecules 42, 43, and 44 to the responsive molecule 46, and a second route is possible from the causative molecule 41 to the connecting molecules 42 and 45 to the responsive molecule 46. Here, for example, if the flow rate value of the connecting molecule 44 is calculated to be zero, the first route is eliminated and the second route is adopted.

[0039] 4(b), in a situation where the first path and the second path are possible, as in FIG. 4(a) (however, the flow value of the connecting molecule 44 is ≠ 0), the flow value of one connecting molecule 43 branching from the connecting molecule 42 is calculated to be 0.2, and the flow value of the other connecting molecule 45 is calculated to be 0.6. Here, if the threshold value for similarity is 0.4, the flow value of the connecting molecule 45 exceeds the upper limit of similarity, so the second path is eliminated and the first path is adopted. Note that if the flow values ​​of the connecting molecules 43 and 45 are both less than 0.4, both the first path and the second path are adopted.

[0040] 5 is a diagram showing an example of a pathway generation method by the pathway generation unit 11. First, as shown in FIG. 5(a), the pathway generation unit 11 places causative molecules upstream and responsive molecules downstream. Then, it sets a start point further upstream of all causative molecules and a finish point further downstream of all responsive molecules.

[0041] Next, as shown in Figure 5(b), the pathway generation unit 11 applies a minimum flow algorithm to the section from the start point to the end point, and generates a pathway by placing connecting molecules between the causative molecules and the responsive molecules based on the connection relationships indicated in the interactome information. At this time, the pathway generation unit 11 calculates the flow value of each molecule so as to satisfy conditions (A) to (C), and performs optimization processing based on the calculated flow values ​​so as to satisfy conditions (1) and (2).

[0042] In addition, when it is specified that a specific molecule be knocked out for a pathway that has already been generated, the pathway generation unit 11 regenerates the pathway using the minimum flow algorithm so that conditions (1) and (2) are satisfied, with the molecule specified for knockout deleted.

[0043] Figure 6 shows an example of regenerating a pathway by knocking out one connecting molecule 60 contained in a pathway that has already been generated. Figure 6(a) shows the pathway generated as in Figure 5(b), in which one connecting molecule 60 is knocked out. The molecule to be knocked out can be specified arbitrarily. For example, it is possible to knock out a molecule that is a target gene for drug discovery.

[0044] Knocking out a molecule 60 is equivalent to forcibly setting the flow value of that molecule 60 to zero. Therefore, as shown in FIG. 6(b), the pathway that passes through the knocked-out molecule 60 is eliminated. When a pathway including the knocked-out molecule 60 disappears, the flow value that was allocated to that pathway is allocated to another pathway. At that time, the pathway generation unit 11 recalculates the flow value of each molecule so as to satisfy conditions (A) to (C), and re-executes the optimization process based on the recalculated flow values ​​so as to satisfy conditions (1) and (2).

[0045] Therefore, in addition to the pathway including the knocked-out molecule 60, a pathway that was adopted before the knockout because it satisfied conditions (1) and (2) may arise, but after the knockout, it no longer satisfies conditions (1) and (2) and therefore disappears. Conversely, a pathway that was not adopted before the knockout because it did not satisfy conditions (1) and (2) may arise, but after the knockout, it satisfies conditions (1) and (2) and therefore is adopted. As a result, the pathway before the knockout shown in Figure 6(a) changes to the pathway shown in Figure 6(c). In Figure 6(c), dotted lines indicate eliminated pathways, and thick lines indicate newly created pathways.

[0046] As explained in detail above, in this embodiment, when generating a pathway based on property information that represents the properties of molecules that act on a disease or symptom, interactome information that represents the connection relationships between molecules, and similarity information that represents the similarity between molecules, the pathway is generated using a minimum flow algorithm so as to satisfy at least two conditions: (1) generating a pathway by connecting only molecules whose flow values ​​are greater than a minimum threshold and less than a maximum threshold, and (2) generating a pathway by connecting only molecules whose flow values ​​are equal to or less than a threshold related to similarity.

[0047] According to this embodiment configured as described above, not only are pathways that pass through molecules with flow values ​​below a minimum threshold or above a maximum threshold eliminated, but pathways that connect molecules with low similarity and pathways that pass through molecules with flow values ​​exceeding a threshold related to similarity are also eliminated. As a result, the entire pathway (intermolecular network) is generated by selecting intermolecular connections based on the magnitude of flow values ​​and similarity from among the intermolecular connections defined by interactome information that represents the connectivity between molecules.

[0048] In this embodiment, the similarity of molecular feature vectors calculated from text data for a certain disease or symptom is used, so that from among the molecular connections defined by the interactome information, molecular connections based on the similarity of molecular feature vectors for the specified disease or symptom are preferentially selected, and a pathway reflecting the intermolecular interactions for the specified disease or symptom is created.

[0049] In other words, the similarity between molecules used in this embodiment indicates the proximity of two molecular feature vectors obtained by analyzing multiple documents (such as literature information). By setting a threshold value for this similarity (e.g., a threshold determined from the similarity) as the upper limit of the flow rate that can pass between two molecules, in the case of a pathway (an important pathway) that requires a certain amount of flow or more, pathways between two molecules with low correlation are eliminated by flow restriction, and a pathway with high correlation is adopted. In other words, if the flow rate value exceeds the upper limit, processing such as replacing it with a molecule with higher similarity or changing the flow rate value of another pathway so that it does not exceed the upper limit is performed. Therefore, pathways with high correlation between two molecules are selected as important pathways, and a connection between two molecules based on literature information on a certain disease is preferentially selected, resulting in the generation of a pathway that is closer to the disease state.

[0050] Therefore, when a specific molecule is knocked out of a pathway that has already been generated and the pathway is regenerated, changes occur not only in the pathway in which the knocked-out molecule existed but also in other pathways (including pathways distant from the pathway in which the knocked-out molecule existed) according to changes in the relationship between the redistributed flow value and the threshold, and an appropriate pathway that reflects the knocked-out state is generated. As described above, according to this embodiment, it is possible to analyze what effect occurs on the entire pathway when a specific molecule is knocked out in a pathway that has already been generated.

[0051] This can be used to predict, for example, the efficacy, resistance, and side effects of drugs for specific diseases. As an example, an analysis unit can be further provided that analyzes the difference between a newly constructed pathway created by knocking out a specific molecule (gene) and the pathway before the knockout, and anticancer drug resistance can be predicted using as an indicator whether the genes that have changed in the pathway before and after the knockout are cancer-related genes (Cancer Hallmarks). The pathway obtained by knocking out a specific gene is thought to be a pathway that shows anticancer drug resistance (treatment resistance) when anticancer drug therapy targeting that specific gene is administered.

[0052] Specifically, in this case, the analysis is performed as follows: First, a disease pathway for the cancer to be analyzed is constructed, and a candidate drug target gene X is selected. Next, the selected gene X is knocked out to construct a new disease pathway. Then, the constituent genes that have changed between the two disease pathways are extracted, and the functions of these constituent genes are analyzed. Here, it is analyzed whether the function of the changed gene corresponds to a Cancer Hallmark (an indicator that characterizes cancer). If the changed gene corresponds to a Cancer Hallmark, this indicates the possibility that new cancer will develop even if the target gene is inhibited, and it is possible to predict anticancer drug resistance.

[0053] In the above embodiment, in addition to conditions (1) and (2), one or both of the following conditions (3) and (4) may be added: (3) The network size of the pathway must be within a predetermined range. The network size may be, for example, the number of molecules included on the path from the causative molecule to the responsive molecule, or the total number of molecules included in the entire pathway. (4) When multiple paths are possible from one causative molecule to one responsive molecule, the shortest path (the path containing the fewest number of molecules) must be adopted. Alternatively, the path with the largest total flow value of each molecule from the causative molecule to the responsive molecule must be adopted.

[0054] In the above embodiment, the flow value of each molecule is a value determined taking into account the similarity between molecules, but the present invention is not limited to this. For example, the flow value of each molecule may be a value calculated independently of the similarity between molecules, and the flow value of each molecule may not be the similarity between molecules.

[0055] Furthermore, the disease pathway or symptom pathway may be generated using, for example, the algorithm described in Japanese Patent No. 6915818.

[0056] Below, we will briefly explain a pathway generation method using the algorithm described in Japanese Patent No. 6915818. Figure 7 is a block diagram showing an example of the functional configuration of a pathway generation device 100 when this algorithm is used. Note that the explanation will be given here using an example of generating a disease pathway.

[0057] The disease feature vector identification unit 101 identifies a feature vector corresponding to a disease name (hereinafter referred to as a disease feature vector). The disease feature vector is data that represents the features of a disease (features that can identify a disease) as a combination of multiple element values. As an example, a vector that represents the contribution of a disease name contained as a word in multiple sentences to each sentence and to what extent is used as the disease feature vector.

[0058] Such a disease feature vector can be calculated, for example, by the feature vector calculation device 200 shown in Fig. 2. The feature vector calculation device 200 inputs sentence data related to a sentence and calculates a disease feature vector that reflects the relationship between the sentence and the words contained therein. That is, in the index value matrix DW calculated as shown in Fig. 3, of the n sets of word index values ​​constituting each column, a word index value group related to a word corresponding to a disease name is identified as the disease feature vector for each disease name.

[0059] The associated molecule inference unit 102 infers a plurality of molecules associated with a disease by inputting the disease feature vector identified by the disease feature vector identification unit 101 into a first trained model pre-stored in the first model storage unit 111. Here, the first trained model is machine-trained so that, when a disease feature vector is input, it outputs information about molecules corresponding to a molecular feature vector similar to the disease feature vector, based on the similarity between the disease feature vector and the molecular feature vector.

[0060] A disease feature vector for a plurality of disease names is calculated, and a molecule feature vector for a plurality of molecule names is calculated using the feature vector calculation device 200. Then, machine learning of a first trained model is performed in advance using these datasets, and the first trained model trained based on the similarity between the disease feature vector and the molecule feature vector is stored in the first model storage unit 111.

[0061] Here, the similarity between the disease feature vector and the molecule feature vector can be evaluated by various methods. For example, a method can be applied in which features are extracted from each of the disease feature vector and the molecule feature vector using a predetermined function, and the similarity of the features is evaluated. Alternatively, the Euclidean distance or cosine similarity between the word index value group of the disease feature vector and the word index value group of the molecule feature vector may be used, or the edit distance may be used.

[0062] The molecular property estimation unit 103 inputs the disease feature vector identified by the disease feature vector identification unit 101 and the molecular feature vector identified for multiple molecules estimated by the related molecule estimation unit 102 into a second trained model stored in the second model storage unit 112, and estimates the probability that each of the multiple molecules estimated to be associated with a disease is causative or responsive as a molecular property acting on the disease.

[0063] Here, the second trained model has been machine-trained to output the probability that the molecular property is causative or responsive when a disease feature vector and a molecular feature vector are input, using a dataset of disease feature vectors, molecular feature vectors, and property information representing the properties of molecules that act on the disease as training data.

[0064] The pathway generation unit 104 uses the molecular properties estimated by the molecular property estimation unit 103 and known interactome information indicating the connection relationships between molecules stored in the knowledge DB storage unit 113 to generate a pathway (intermolecular network) that represents intermolecular interactions as a route diagram, by arranging the causative molecule upstream and the responsive molecule downstream for multiple molecules whose association with a disease has been estimated by the related molecule estimation unit 102, and by placing other connecting molecules between the causative molecule and the responsive molecule, reflecting the connection relationships indicated by the interactome information.

[0065] At this time, the pathway generation unit 104 calculates the flow rate values ​​of each molecule so as to satisfy conditions (A) to (C), and generates a pathway using a minimum flow algorithm based on the calculated flow rate values ​​so as to satisfy conditions (1) and (2), for example, so that causative molecules whose probability value estimated by the molecular property estimation unit 103 to be causative is greater than the first threshold value Th1 are placed upstream of the pathway, responsive molecules whose probability value is less than the second threshold value Th2 (Th1 > Th2) are placed downstream of the pathway, and connecting molecules whose probability value is greater than the second threshold value Th2 and less than the first threshold value Th1 are placed between the causative molecule and the responsive molecule.

[0066] Here, the similarity used in condition (B) when calculating the flow rate value may be, for example, a value of the similarity between the disease feature vector and the molecular feature vector identified by the associated molecule inference unit 102. Furthermore, a threshold value related to the similarity between the disease feature vector and the molecular feature vector may be used as a threshold value related to the similarity used in condition (2).

[0067] Furthermore, the above-described embodiments are merely examples of specific embodiments for carrying out the present invention, and the technical scope of the present invention should not be construed as being limited thereby. In other words, the present invention can be carried out in various forms without departing from the gist or main characteristics thereof.

[0068] 10, 100 Pathway generation device 11 Pathway generation unit 12 Connection information storage unit 13 Property information storage unit 14 Similarity information storage unit

Claims

1. A pathway generation device comprising a pathway generation unit that generates a pathway that represents intermolecular interactions as a path diagram based on property information that represents the properties of molecules that act on a disease or symptom, connection information that represents the connection relationships between molecules, and similarity information that represents the similarity between molecules, so that causative molecules represented by the property information are located upstream and responsive molecules are located downstream, and also reflects the connection relationships represented by the connection information and makes it easier for molecules with high similarity represented by the similarity information to be connected to each other, wherein the pathway generation unit generates the pathway using a minimum flow algorithm to satisfy at least two of the following conditions: (1) generating a pathway by connecting only molecules whose flow values ​​are greater than a minimum threshold and less than a maximum threshold, and (2) generating a pathway by connecting only molecules whose flow values ​​are equal to or less than the threshold related to the similarity.

2. The pathway generation device of claim 1, characterized in that the pathway generation unit determines the flow value of each molecule based on the similarity between the molecules, and applies conditions (1) and (2) by setting the maximum threshold value of the flow value and the threshold value related to the similarity to be the same value.

3. The pathway generation device of claim 1 or 2, characterized in that, when it is specified that a specific molecule be knocked out of a pathway that has already been generated, the pathway generation unit regenerates the pathway using a minimum flow algorithm so as to satisfy at least two of the above conditions, with the molecule specified to be knocked out deleted.

4. The pathway generation device as described in claim 3, further comprising an analysis unit that analyzes molecules that have changed in the pathway regenerated with the specific molecule deleted from the pathway generated before the specific molecule was deleted.

5. A pathway generation method comprising the steps of: a pathway generation unit of a computer generates a pathway that represents intermolecular interactions as a path diagram based on property information that represents the properties of molecules that act on a disease or symptom, connection information that represents the connection relationships between molecules, and similarity information that represents the similarity between molecules, such that causative molecules represented by the property information are located upstream and responsive molecules are located downstream, while reflecting the connection relationships represented by the connection information and making it easier for molecules with high similarity represented by the similarity information to be connected to each other; wherein the pathway generation unit generates the pathway using a minimum flow algorithm to satisfy at least two of the following conditions: (1) generating a pathway by connecting only molecules whose flow values ​​are greater than a minimum threshold and less than a maximum threshold, and (2) generating a pathway by connecting only molecules whose flow values ​​are equal to or less than the threshold related to the similarity.

6. A program for causing a computer to function as a pathway generation means for generating a pathway that represents intermolecular interactions as a path diagram, based on property information that represents the properties of molecules that act on a disease or symptom, connection information that represents the connection relationships between molecules, and similarity information that represents the similarity between molecules, so that causative molecules represented by the property information are located upstream and responsive molecules are located downstream, while reflecting the connection relationships represented by the connection information and making it easier for molecules with high similarity represented by the similarity information to be connected to each other, wherein the pathway generation means generates the pathway using a minimum flow algorithm to satisfy at least two of the following conditions: (1) generating a pathway by connecting only molecules whose flow values ​​are greater than a minimum threshold and less than a maximum threshold, and (2) generating a pathway by connecting only molecules whose flow values ​​are equal to or less than the threshold related to the similarity.