Low molecular weight compound searching device, low molecular weight compound searching method, and program

Abstract

To provide a low molecular weight compound searching device that can reduce the time required for a search of a low molecular weight compound.SOLUTION: A low molecular weight compound searching device 2 comprises: a restriction information acquisition unit 202 that acquires chemical structure restriction information B1 indicating restrictions on a chemical structure required for a compound; a first chemical structure selection unit 203 that selects, from a plurality of first low molecular weight compounds, second low molecular weight compounds having a chemical structure satisfying the restrictions indicated by the chemical structure restriction information B1; and a regression calculation unit 207 that, on the basis of the regression model generated on the basis of the chemical structure of a low molecular weight compound and the physical property value of the low molecular weight compound, calculates, for each of the second low molecular weight compounds, an estimate of the physical property value of the second low molecular weight compound from the chemical structure of the second low molecular weight compound.SELECTED DRAWING: Figure 2

Description

[Technical field]

[0001] The present invention relates to an apparatus for searching for small molecular compounds, a method for searching for small molecular compounds, and a program. [Background technology]

[0002] Compounds having desired physical properties are searched for based on information science. For example, a search method for searching low molecular weight compounds based on a regression model is known (Patent Document 1). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2023-54446 A Summary of the Invention [Problem to be solved by the invention]

[0004] The compound space, which is a collection of chemical structures of compounds, is very wide, and the number of chemical structures to be obtained when searching for low molecular weight compounds is very large. In the searching method described in Patent Document 1, the chemical structures are directly inputted into a regression model to calculate estimated values ​​and the variance of the estimated values. However, when a large number of chemical structures are directly inputted into the regression model, it takes a long time to calculate the estimated values ​​because of the large number of chemical structures. There is a demand for shortening the time required for searching for small molecule compounds.

[0005] The present invention has been made in consideration of the above points, and provides a small molecular compound searching device, a small molecular compound searching method, and a program that can shorten the time required for searching for small molecular compounds. [Means for solving the problem]

[0006] In order to solve the above problems, the present invention has the following configuration. [1] A small molecule compound searching device comprising: a constraint information acquisition unit that acquires chemical structure constraint information indicating constraints on chemical structure required for a compound; a first chemical structure selection unit that selects, from a plurality of first small molecule compounds, second small molecule compounds having a chemical structure that satisfies the constraints indicated by the chemical structure constraint information acquired by the constraint information acquisition unit; and a regression calculation unit that calculates, for each of the second small molecule compounds selected by the first chemical structure selection unit, an estimate of the physical property value of the second small molecule compound from the chemical structure of the second small molecule compound, based on a regression model generated based on the chemical structure of the low molecular weight compound and the physical property values ​​of the low molecular weight compound.

[0007] [2] The small molecule compound screening device according to [1], further comprising a second chemical structure selection unit that selects a third small molecule compound that satisfies a target physical property value from the second small molecule compounds selected by the first chemical structure selection unit based on the estimated value calculated by the regression calculation unit.

[0008] [3] The small molecule compound searching device described in [1], wherein the regression calculation unit calculates, for each of the second small molecule compounds selected by the first chemical structure selection unit, a probability distribution of physical property values ​​of the second small molecule compound from the chemical structure of the second small molecule compound based on the regression model, and calculates an acquisition function from the calculated probability distribution.

[0009] [4] The small molecule compound discovery device according to [3], wherein the regression model is a regression model based on any one of Gaussian process regression, Bayesian ridge regression, ensemble regression, deep learning, or gradient boosting-based regression.

[0010] [5] The small molecule compound screening apparatus according to [1], further comprising a chemical structure generation unit that generates a structural formula showing a chemical structure of the first small molecule compound for each of the plurality of first small molecule compounds.

[0011] [6] The small molecule compound discovery device described in [1], wherein the plurality of first small molecule compounds include at least one of compounds selected by a human, compounds extracted from a database, and compounds designed by a specific algorithm.

[0012] [7] The small molecule compound discovery device according to [1], wherein the physical property values ​​used to generate the regression model are at least one of actual measured values ​​of the physical property values ​​of a small molecule compound, or calculated values ​​calculated or predicted by a specific algorithm.

[0013] [8] The small molecule compound searching device according to [1], further comprising a constraint condition generation unit that generates the chemical structure constraint information based on constraint generation information indicating a desired reaction used in the synthesis of a small molecule compound or a physical property of the small molecule compound.

[0014] [9] The small molecule compound discovery device described in [1], wherein the constraints indicated by the chemical structure constraint information include a first constraint and a second constraint that applies to more chemical structures of small molecule compounds than the first constraint, and the first chemical structure selection unit selects second small molecule compounds having chemical structures that satisfy the second constraint, and then selects second small molecule compounds having chemical structures that satisfy the first constraint from the selected second small molecule compounds.

[0015]

[10] The low molecular weight compound screening device according to [1], further comprising a character string conversion unit for converting a chemical structure of the low molecular weight compound into a character string.

[0016]

[11] A method for discovering low molecular weight compounds, comprising: acquiring chemical structure constraint information indicating constraints on a chemical structure required for a compound; selecting, from a plurality of first low molecular weight compounds, second low molecular weight compounds having a chemical structure that satisfies the constraints indicated by the acquired chemical structure constraint information; and calculating, for each of the selected second low molecular weight compounds, an estimate of the physical property value of the second low molecular weight compound from the chemical structure of the second low molecular weight compound, based on a regression model generated based on the chemical structure of the low molecular weight compound and the physical property values ​​of the low molecular weight compound.

[0017]

[12] A program for causing a computer to execute the steps of acquiring chemical structure constraint information indicating constraints on a chemical structure required for a compound, selecting from a plurality of first low molecular weight compounds second low molecular weight compounds having a chemical structure that satisfies the constraints indicated by the acquired chemical structure constraint information, and calculating, for each of the selected second low molecular weight compounds, an estimate of the physical property value of the second low molecular weight compound from the chemical structure of the second low molecular weight compound based on a regression model generated based on the chemical structure of the low molecular weight compound and the physical property values ​​of the low molecular weight compound. Effect of the Invention

[0018] According to the present invention, the time required for searching for low molecular weight compounds can be reduced. [Brief description of the drawings]

[0019] [Figure 1] FIG. 1 is a diagram showing an example of the configuration of a small molecule compound screening system 1 according to a first embodiment of the present invention. [Diagram 2] FIG. 1 is a diagram showing an example of the functional configuration of a small molecule compound screening apparatus 2 according to a first embodiment of the present invention. [Diagram 3] FIG. 2 is a diagram showing an example of a structure screening process according to the first embodiment of the present invention. [Figure 4] FIG. 4 is a diagram illustrating an example of a learning process according to the first embodiment of the present invention. [Diagram 5] FIG. 2 is a diagram showing an example of a property screening process according to the first embodiment of the present invention. [Figure 6] FIG. 1 is a diagram showing an example of the functional configuration of a small molecule compound screening apparatus 2a according to a second embodiment of the present invention. [Figure 7] FIG. 11 is a diagram showing an example of a structure screening process according to the second embodiment of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0020] (First embodiment) Hereinafter, an embodiment of the present invention will be described in detail with reference to the drawings. Fig. 1 is a diagram showing an example of the configuration of a small molecule compound discovery system 1 according to this embodiment. The small molecule compound discovery system 1 includes a small molecule compound discovery device 2, an input device 3, and a presentation device 4.

[0021] The low molecular compound searching apparatus 2 searches for a low molecular compound that satisfies a target physical property value from among low molecular compounds. Here, the low molecular compound searching apparatus 2 selects, as the chemical structure of the low molecular compound to be searched for, a chemical structure of a low molecular compound that satisfies the chemical structure constraints required of the compound from among the chemical structures of low molecular compounds. In other words, the low molecular compound searching apparatus 2 performs screening of the chemical structures of low molecular compounds based on the chemical structure constraints. The low molecular compound searching apparatus 2 is, for example, a computer such as a personal computer (PC) or a server.

[0022] The input device 3 inputs various data to the small molecule compound searching apparatus 2. The various data input by the input device 3 includes chemical structure constraint information B1. The chemical structure constraint information B1 indicates constraints on the chemical structure required for a compound. The chemical structure constraint information B1 is used by the small molecule compound searching apparatus 2 to perform screening on chemical structures. The chemical structure constraint information B1 will be described later.

[0023] The input device 3 is, for example, a keyboard, a mouse, or a touch panel. The input device 3 may be an external storage device that stores various data to be input to the small molecular compound screening apparatus 2. The input device 3 may also be a terminal device that communicates with the small molecular compound screening apparatus 2.

[0024] The presentation device 4 presents various data used for processing by the small molecular compound screening apparatus 2 and the search results by the small molecular compound screening apparatus 2. The presentation device 4 is, for example, a display. The presentation device 4 may be an external server that communicates with the small molecular compound screening apparatus 2.

[0025] In this embodiment, a low molecular weight compound is a compound having a molecular weight of 10,000 or less. In the following description, a low molecular weight compound before screening for chemical structure by the low molecular weight compound searching device 2 is referred to as a first low molecular weight compound. Among the first low molecular weight compounds, a low molecular weight compound that satisfies the constraints indicated by the chemical structure constraint information B1 is referred to as a second low molecular weight compound. Among the second low molecular weight compounds, a low molecular weight compound that satisfies the target physical property values ​​is referred to as a third low molecular weight compound.

[0026] [Functional configuration of the low molecular compound discovery device 2] 2 is a diagram showing an example of the functional configuration of the small molecule compound screening apparatus 2 according to this embodiment. The small molecule compound screening apparatus 2 includes a control unit 20 and a storage unit .

[0027] The control unit 20 includes, for example, a CPU (Central Processing Unit) and performs various calculations and information transmission and reception. The control unit 20 includes a chemical structure generation unit 200, a structure screening unit 201, a property screening unit 204, a character string conversion unit 209, and an output unit 210. Each of these functional units included in the control unit 20 is realized, for example, by the CPU loading a program read from a ROM (Read Only Memory) into a RAM (Random Access Memory) and executing processing according to the program. The ROM and RAM are included in the storage unit 21.

[0028] The chemical structure generation unit 200 generates a structural formula showing the chemical structure of the first low molecular weight compound for each of the first low molecular weight compounds. The chemical structure generation unit 200 stores information showing the generated structural formula as first low molecular weight compound structural formula information A1 in the storage unit 21. The first low molecular weight compound structural formula information A1 is information showing a structural formula showing the chemical structure of the first low molecular weight compound. Note that information showing the first low molecular weight compounds for which the chemical structure generation unit 200 generates structural formulas is stored in advance in the storage unit 21 (not shown).

[0029] The structure screening unit 201 performs screening of the chemical structures of a plurality of first low molecular weight compounds. The structure screening unit 201 includes a constraint information acquisition unit 202 and a first chemical structure selection unit 203.

[0030] The constraint information acquisition unit 202 acquires chemical structure constraint information B1. The chemical structure constraint information B1 is acquired by acquiring the chemical structure constraint information B1 input from the input device 3. The chemical structure constraint information B1 is information indicating constraints on a chemical structure required for a compound. The chemical structure constraint information B1 may indicate one or more constraints. The constraint information acquisition unit 202 stores the acquired chemical structure constraint information B1 in the storage unit 21.

[0031] The chemical structure constraints required for the compound include, for example, not containing a carboxylic acid, containing an ether structure, not containing five or more halide atoms, having a five- or six-membered ring structure, not containing bonds between heteroatoms, or containing a heterocycle.

[0032] The first chemical structure selection unit 203 selects, from among a plurality of first low molecular weight compounds, a second low molecular weight compound having a chemical structure that satisfies the constraints indicated by the chemical structure constraint information B1 acquired by the constraint information acquisition unit 202. The first chemical structure selection unit 203 stores the selection result as second low molecular weight compound structural formula information C1 in the storage unit 21. The second low molecular weight compound structural formula information C1 is information indicating the second low molecular weight compound selected by the first chemical structure selection unit 203 by the structural formula of the second low molecular weight compound.

[0033] The physical property screening unit 204 performs screening of the physical properties of a plurality of second low molecular weight compounds. The physical property screening unit 204 includes a feature amount calculation unit 205, a learning unit 206, a regression calculation unit 207, and a second chemical structure selection unit 208.

[0034] The feature amount calculation unit 205 calculates feature amounts from the low molecular weight compound physical property data D1. The low molecular weight compound physical property data D1 is data indicating, for a low molecular weight compound, the chemical structure of the low molecular weight compound and the physical property values ​​of the low molecular weight compound. The low molecular weight compound is, for example, a low molecular weight compound extracted from a database of compounds provided outside the low molecular weight compound discovery system 1. The physical property value is at least one of an actual measured value or a calculated value calculated or predicted by a specific algorithm.

[0035] The low molecular weight compound physical property data D1 has columns for each item, such as information for identifying a low molecular weight compound, chemical structure, and physical property value. The information for identifying a low molecular weight compound is, for example, a name, an identifier, etc. The low molecular weight compound physical property data D1 is two-dimensional tabular data consisting of rows and columns in which the chemical structure and physical property values ​​are stored for each low molecular weight compound. The low molecular weight compound physical property data D1 is stored in advance in the storage unit 21.

[0036] The feature amount calculation section 205 stores the calculated feature amount as feature amount information E1 in the storage section 21. The feature amount information E1 is information indicating the feature amount calculated by the feature amount calculation section 205 from the low molecular weight compound physical property data D1.

[0037] The learning unit 206 generates a regression model based on the feature amount information E1. In other words, the learning unit 206 generates a regression model based on the chemical structure of the low molecular weight compound and the physical property value of the low molecular weight compound. The learning unit 206 stores the generated regression model in the storage unit 21 as a trained model F1.

[0038] The regression calculation unit 207 calculates, for each of the second low molecular weight compounds selected by the first chemical structure selection unit 203, an estimate of the physical property value of the second low molecular weight compound from the chemical structure of the second low molecular weight compound based on the regression model. The regression model is the trained model F1. The second low molecular weight compounds selected by the first chemical structure selection unit 203 are represented by the second low molecular weight compound structural formula information C1 as described above. The regression calculation unit 207 stores the estimation result in the storage unit 21 as predicted physical property value information G1. Note that calculating an estimated value of a physical property value is also referred to as estimating a physical property value or predicting a physical property value.

[0039] The second chemical structure selection unit 208 selects a third low molecular weight compound that satisfies a target physical property value from among the second low molecular weight compounds selected by the first chemical structure selection unit 203 based on the estimated value calculated by the regression calculation unit 207. The target physical property value is a physical property value desired by a user of the low molecular weight compound searching system 1. The target physical property value is input from the input device 3, for example, and stored in the storage unit 21. The target physical property value may be stored in advance in the storage unit 21. The second chemical structure selection unit 208 stores the selection result in the storage unit 21 as third low molecular weight compound structural formula information H1.

[0040] The character string conversion unit 209 converts the chemical structure of a low molecular weight compound into a character string. The character string is, for example, a character string in the Simplified Molecular Input Line Entry System (SMILES). Note that the character string may be in a notation method other than SMILES, such as a MOL file or SDF.

[0041] In the processing performed by the control unit 20, the chemical structure is converted into a character string and then processed. Before the processing based on the chemical structure among the processing performed by the control unit 20 is performed, the character string conversion unit 209 converts the chemical structure into a character string. Each of the first low molecular weight compound structural formula information A1, the chemical structure constraint information B1, the second low molecular weight compound structural formula information C1, and the low molecular weight compound physical property data D1 includes a chemical structure, but the chemical structure may be included in this information after being converted into a character string in advance by the character string conversion unit 209.

[0042] The output unit 210 outputs the third low molecular weight compound structural formula information H1 as a search result by the low molecular weight compound searching device 2. The output unit 210 outputs the third low molecular weight compound structural formula information H1 to the presentation device 4. The output unit 210 may cause an external server to output the third low molecular weight compound structural formula information H1.

[0043] The storage unit 21 stores various information. For example, the storage unit 21 stores first low molecular weight compound structural formula information A1, chemical structure constraint information B1, second low molecular weight compound structural formula information C1, low molecular weight compound physical property data D1, feature amount information E1, trained model F1, predicted physical property value information G1, and third low molecular weight compound structural formula information H1. The storage unit 21 is configured using a storage device such as a magnetic hard disk device or a semiconductor storage device.

[0044] The small molecule compound screening apparatus 2 may be realized as a virtual server. Each functional unit of the small molecule compound screening apparatus 2 may be distributed among a plurality of servers. The small molecule compound screening apparatus 2 may be realized as a cloud server.

[0045] Here, the processes performed by the small molecule compound searching apparatus 2 include a structure screening process, a learning process, and a physical property screening process. The structure screening process, the learning process, and the physical property screening process will be described with reference to Fig. 3, Fig. 4, and Fig. 5.

[0046] [Structure screening process] 3 is a diagram showing an example of the structure screening process according to this embodiment. The structure screening process is a process for generating second low molecular weight compound structural formula information C1, and is executed by the control unit 20. The structure screening process is executed, for example, before the learning process. Note that the structure screening process may be executed after the learning process. The structure screening process and the learning process may be executed in parallel.

[0047] Step S10: The chemical structure generation unit 200 generates a structural formula showing the chemical structure of each of the first low molecular weight compounds for each of the first low molecular weight compounds. The chemical structure generation unit 200 stores information showing the generated structural formula in the storage unit 21 as first low molecular weight compound structural formula information A1.

[0048] As described above, information indicating a plurality of first low molecular weight compounds for which the chemical structure generation unit 200 generates a structural formula is stored in advance in the storage unit 21. The plurality of first low molecular weight compounds for which the chemical structure generation unit 200 generates a structural formula include at least one of compounds selected by a person, compounds extracted from a database, and compounds designed by a specific algorithm.

[0049] The person here refers to a user of the small molecular compound discovery system 1. The database refers to, for example, a database of compounds provided outside the small molecular compound discovery system 1. Examples of the specific algorithm include an algorithm that fragments a molecule into substructures, an algorithm that fragments a molecule into substructures and then connects these substructures, and an algorithm that lists any substructures and then connects these substructures.

[0050] Step S20: The chemical structure constraint information B1 is acquired from the input device 3. For example, a user of the small molecule compound searching system 1 operates a keyboard, a mouse, a touch panel, or the like as the input device 3 to input constraints on the chemical structure required for a compound to the small molecule compound searching apparatus 2. The constraint information acquisition unit 202 stores the acquired chemical structure constraint information B1 in the storage unit 21.

[0051] Step S30: The first chemical structure selection unit 203 selects, from among the multiple first low molecular weight compounds, a second low molecular weight compound having a chemical structure that satisfies the constraints indicated by the chemical structure constraint information B1 acquired by the constraint information acquisition unit 202, based on the structural formula generated by the chemical structure generation unit 200.

[0052] The structural formula generated by the chemical structure generation unit 200 is indicated by first low molecular weight compound structural formula information A1. The first chemical structure selection unit 203 performs a selection process based on a character string obtained by converting the structural formula indicated by the first low molecular weight compound structural formula information A1 by the character string conversion unit 209.

[0053] Here, when the chemical structure constraint information B1 indicates multiple constraints, the first chemical structure selection unit 203 selects second low molecular weight compounds having chemical structures that satisfy the constraints from among the multiple constraints, starting with the constraint that applies to as many chemical structures of the low molecular weight compounds as possible.

[0054] Therefore, the constraints indicated by the chemical structure constraint information B1 include a first constraint and a second constraint that applies to more chemical structures of low molecular weight compounds than the first constraint. After selecting second low molecular weight compounds having chemical structures that satisfy the second constraint, the first chemical structure selection unit 203 selects second low molecular weight compounds having chemical structures that satisfy the first constraint from the selected second low molecular weight compounds. The first constraint and the second constraint are determined by the compound distribution of the first low molecular weight compound structural formula information A1 to be selected. For example, assume that the first low molecular weight compound structural formula information A1 contains many structural formulas containing carboxylic acids and not many structural formulas containing esters. Also assume that the structures not containing carboxylic acids and the structures not containing esters are each constraints. In this case, the constraint of the structures not containing carboxylic acids is an example of the second constraint, and the structures not containing esters is an example of the first constraint.

[0055] The order of the constraints used by the first chemical structure selection unit 203 for selection is not limited to the above order. The first chemical structure selection unit 203 may select second low molecular weight compounds having chemical structures that satisfy the constraints among the multiple constraints in the order specified in the chemical structure constraint information B1 or in a random order. The order of the constraints used by the first chemical structure selection unit 203 for selection may be specified by the user of the low molecular weight compound searching system 1.

[0056] The first chemical structure selection section 203 stores the selection result in the storage section 21 as second low molecular weight compound structural formula information C1. With this, the control unit 20 ends the structure screening process.

[0057] [Learning process] 4 is a diagram showing an example of the learning process according to the present embodiment. The learning process is a process for generating a regression model, and is executed by the control unit 20.

[0058] Step S110: The feature amount calculation unit 205 calculates feature amounts from the low molecular weight compound physical property data D1. The feature amount calculation unit 205 stores the calculated feature amounts in the storage unit 21 as feature amount information E1.

[0059] Here, the feature calculation unit 205 calculates the feature from the chemical structure included in the low molecular weight compound physical property data D1. The feature calculation unit 205 calculates the feature from a character string into which the chemical structure included in the low molecular weight compound physical property data D1 is converted by the character string conversion unit 209.

[0060] The feature amount calculated from the chemical structure is, for example, a molecular descriptor. A molecular descriptor is a feature amount such as a molecular characteristic converted into a number. The molecular descriptor includes, for example, one or more of the molecular weight, the number of atoms, the number of heavy atoms, the number of halogen atoms, the number of rings, the number of double bonds, the number of triple bonds, the number of rotatable bonds, the molecular size, and the topological polar surface area (TPSA).

[0061] As described above, the low molecular weight compounds included in the low molecular weight compound property data D1 used to generate the regression model are low molecular weight compounds extracted from a compound database. The low molecular weight compounds used to generate the regression model may include a first low molecular weight compound. The low molecular weight compounds used to generate the regression model may also include a second low molecular weight compound selected by structural screening performed by the structural screening unit 201.

[0062] Step S120: The learning unit 206 generates a regression model based on the feature amount information E1. In other words, the learning unit 206 generates a regression model based on the chemical structure of a low molecular weight compound and the physical property values ​​of the low molecular weight compound.

[0063] Here, the regression model is, for example, a regression model based on Gaussian process regression. Note that the regression model may be a regression model based on any of Bayesian ridge regression, ensemble regression, deep learning, and gradient boosting-based regression. Furthermore, the physical property values ​​used to generate the regression model are at least one of the actual measured values ​​of the physical property values ​​of the low molecular weight compound, or calculated or predicted values ​​calculated by a specific algorithm.

[0064] The learning unit 206 stores the generated regression model in the memory unit 21 as a learned model F1. With this, the control unit 20 ends the learning process.

[0065] 5 is a diagram showing an example of the property screening process according to this embodiment. The property screening process is a process for generating the third low molecular weight compound structural formula information H1, and is executed by the control unit 20. The property screening process is executed after the structure screening process and the learning process.

[0066] Step S210: The regression calculation unit 207 calculates, based on the regression model, estimates of the physical property values ​​of each of the second low molecular weight compounds selected by the first chemical structure selection unit 203 from the chemical structures of the second low molecular weight compounds.

[0067] In this embodiment, as an example, the regression calculation unit 207 calculates a probability distribution of the physical property values ​​of the second low molecular weight compound from the chemical structure of the second low molecular weight compound for each second low molecular weight compound selected by the first chemical structure selection unit 203 based on a regression model. As described above, in this embodiment, the regression model is a regression model based on Gaussian process regression. The probability distribution of the physical property value is the average value of the probability distribution as an estimated value of the physical property value, and the standard deviation of the probability distribution as the variance of the estimated value. Furthermore, the regression calculation unit 207 calculates an acquisition function from the calculated probability distribution.

[0068] The acquisition function is a function for determining a search point in Gaussian process regression. In this embodiment, the acquisition function is, for example, EI (Expected Improvement). As another example, the acquisition function may be any one of PI (Probability of Improvement), UCB (Upper Confidence Bound), LCB (Lower Confidence Bound), etc.

[0069] Step S220: The second chemical structure selection unit 208 selects a third low molecular weight compound that satisfies the target physical property value from among the second low molecular weight compounds selected by the first chemical structure selection unit 203 based on the estimated value calculated by the regression calculation unit 207.

[0070] Here, the second chemical structure selection unit 208 determines the low molecular weight compound for which the acquisition function calculated by the regression calculation unit 207 is maximum or minimum. In this embodiment, the acquisition function is EI. The second chemical structure selection unit 208 determines the low molecular weight compound for which EI is maximum or minimum from among the second low molecular weight compounds as the third low molecular weight compound satisfying the target physical property value. Note that the second chemical structure selection unit 208 may determine a predetermined number of low molecular weight compounds in descending order of EI from among the second low molecular weight compounds as the third low molecular weight compound satisfying the target physical property value. The second chemical structure selection section 208 stores the selection result in the storage section 21 as third low molecular weight compound structural formula information H1.

[0071] Step S230: The output unit 210 outputs the third low molecular weight compound structural formula information H1 as a search result by the low molecular weight compound searching device 2. The output unit 210 outputs the third low molecular weight compound structural formula information H1 to the presentation device 4. With this, the control unit 20 ends the property screening process.

[0072] In this embodiment, an example has been described in which the first low molecular compound structural formula information A1 is generated by the chemical structure generation unit 200, but the present invention is not limited to this. The low molecular compound searching apparatus 2 may acquire the first low molecular compound structural formula information A1 from an external device (such as a database server) separate from the low molecular compound searching apparatus 2. The first low molecular compound structural formula information A1 may be stored in advance in the storage unit 21. In such cases, the chemical structure generation unit 200 may be omitted from the configuration of the low molecular compound searching apparatus 2.

[0073] In addition, in the present embodiment, an example in which the trained model F1 is generated by the training unit 206 has been described, but the present invention is not limited to this. The trained model F1 may be generated by an external device (such as a PC or a server) separate from the small molecular compound screening apparatus 2. In this case, the small molecular compound screening apparatus 2 acquires the trained model F1 from the external device. The trained model F1 may also be stored in advance in the storage unit 21. In such cases, the feature calculation unit 205 and the training unit 206 may be omitted from the configuration of the small molecular compound screening apparatus 2.

[0074] In addition, in this embodiment, an example of a case where a third low molecular weight compound satisfying a target physical property value is selected from the second low molecular weight compounds based on the estimated value calculated by the regression calculation unit 207 has been described, but the present invention is not limited to this. The low molecular weight compound searching device 2 may not select a third low molecular weight compound. In other words, the low molecular weight compound searching device 2 may calculate an estimated value of the physical property value of the second low molecular weight compound from the chemical structure of the second low molecular weight compound, store the calculation result in the storage unit 21 as predicted physical property value information G1, and then terminate the process. In this case, the predicted physical property value information G1 may be output from the low molecular weight compound searching device 2 to an external device (such as a PC or a server) separate from the low molecular weight compound searching device 2, and the third low molecular weight compound satisfying the target physical property value may be selected from the second low molecular weight compounds by the external device. In this case, the second chemical structure selection unit 208 may be omitted from the configuration of the low molecular weight compound searching device 2.

[0075] In the present embodiment, an example of a case where the chemical structure of a low molecular weight compound is converted into a character string by the character string conversion unit 209 has been described, but the present invention is not limited to this. The information including the chemical structure used in each process executed by the low molecular weight compound searching apparatus 2 may include the chemical structure as a character string in advance. In this case, the character string conversion unit 209 may be omitted from the configuration of the low molecular weight compound searching apparatus 2.

[0076] As described above, the small molecule compound searching apparatus 2 according to this embodiment includes the constraint information acquisition section 202 , the first chemical structure selection section 203 , and the regression calculation section 207 . The constraint information acquisition unit 202 acquires chemical structure constraint information B1 that indicates constraints on the chemical structure required for a compound. The first chemical structure selection unit 203 selects, from among a plurality of first low molecular weight compounds, a second low molecular weight compound having a chemical structure that satisfies the constraints indicated by the chemical structure constraint information B1 acquired by the constraint information acquisition unit 202. The regression calculation unit 207 calculates, for each of the second low molecular weight compounds selected by the first chemical structure selection unit 203, an estimate of the physical property value of the second low molecular weight compound from the chemical structure of the second low molecular weight compound, based on a regression model (in this embodiment, the trained model F1) generated based on the chemical structure of the low molecular weight compound and the physical property value of the low molecular weight compound.

[0077] With this configuration, the small molecule compound searching apparatus 2 of this embodiment can select a second small molecule compound having a chemical structure that satisfies the chemical structure constraints required of a compound from among a plurality of first small molecule compounds before calculating an estimated value of a physical property value, thereby reducing the time required for searching for small molecule compounds.

[0078] The number of possible organic molecules is very large, and even if we limit it to low molecular weight compounds, it is said that there are 10^60 compounds, so simply generating structures requires handling a large number of compounds, which takes a lot of time. For example, there are 1,168 possible chemical structures with the molecular formula C5H8O2. However, when we look at these chemical structures, we see that they are diverse and include many that are not suitable for the purpose.

[0079] Therefore, in the low molecular compound searching device 2 according to this embodiment, the number of chemical structures can be reduced by screening with restrictions on the chemical structures according to the purpose. For example, if screening is performed with the requirement that the chemical structures must have a five-membered or six-membered ring structure, the number of chemical structures can be reduced to 119. By screening for chemical structures, the number of compounds can be reduced to one-tenth in advance, and screening for physical properties can be performed quickly. Furthermore, it is possible to prevent the selection of compounds that are unsuitable for the target compound based on the constraints of the chemical structure required of the compound in the first place.

[0080] Furthermore, the small molecular weight compound screening apparatus 2 according to this embodiment may further include a second chemical structure selection unit 208 that selects a third small molecular weight compound that satisfies a target physical property value from among the second small molecular weight compounds selected by the first chemical structure selection unit 203, based on the estimated value calculated by the regression calculation unit 207. With this configuration, the small molecule compound screening apparatus 2 according to this embodiment can select a third small molecule compound that satisfies a target physical property value from among the second small molecules.

[0081] Furthermore, in the small molecular weight compound searching apparatus 2 according to this embodiment, the regression calculation section 207 may calculate, for each of the second small molecular weight compounds selected by the first chemical structure selection section 203, a probability distribution of the physical property values ​​of the second small molecular weight compound from the chemical structure of the second small molecular weight compound based on a regression model, and calculate an acquisition function from the calculated probability distribution. Here, when the probability distribution of physical properties is calculated using a regression model based on Gaussian process regression, it often takes more time than estimation (prediction) using general machine learning. With the above configuration, in the small molecule compound searching device 2 according to this embodiment, the second small molecule compound is pre-selected by a structure screening process, so that the number of chemical structures used to calculate the probability distribution of physical properties can be reduced. Therefore, the small molecule compound searching device 2 according to this embodiment is suitable for calculating the probability distribution of physical properties using a regression model based on Gaussian process regression.

[0082] Furthermore, in the small molecule compound screening apparatus 2 according to this embodiment, the regression model may be a regression model based on any of Gaussian process regression, Bayesian ridge regression, ensemble regression, deep learning, and gradient boosting-based regression. With this configuration, in the small molecular weight compound screening apparatus 2 of this embodiment, even when a regression model based on any of Gaussian process regression, Bayesian ridge regression, ensemble regression, deep learning, or regression based on gradient boosting is used, the second small molecular weight compound can be pre-selected by the structure screening process to reduce the number of chemical structures used to calculate estimated values ​​of the physical property values ​​of the second small molecular weight compound, thereby shortening the time required for searching for small molecular weight compounds.

[0083] Moreover, the small molecule compound screening apparatus 2 according to this embodiment may further comprise a chemical structure generation section 200 that generates, for each of a plurality of first small molecule compounds, a structural formula showing the chemical structure of the first small molecule compound. With this configuration, the small molecule compound screening apparatus 2 according to this embodiment can generate a structural formula showing the chemical structure of the first small molecule compound, and therefore there is no need to obtain information showing the structural formula from an external device.

[0084] Furthermore, in the small molecule compound screening apparatus 2 of this embodiment, the multiple first small molecule compounds may include at least one of compounds selected by a person, compounds extracted from a database, and compounds designed by a specific algorithm. With this configuration, in the small molecule compound searching apparatus 2 of this embodiment, even if there are many chemical structures of compounds selected by a person, compounds extracted from a database, or compounds designed by a specific algorithm, the second small molecule compound can be pre-selected by a structure screening process, thereby reducing the time required for searching for small molecule compounds.

[0085] Furthermore, in the small molecular weight compound searching apparatus 2 of this embodiment, the physical property values ​​used to generate the regression model (in this embodiment, the trained model F1) may be at least one of the actual measured values ​​of the physical property values ​​of the small molecular weight compound, or calculated values ​​calculated or predicted by a specific algorithm. With this configuration, in the small molecular compound searching apparatus 2 of this embodiment, when actual measured values ​​can be obtained, actual measured values ​​are used as physical property values ​​to be used when generating a regression model, and when actual measured values ​​cannot be obtained, calculated values ​​are used, and thus physical property values ​​can be obtained in various ways.

[0086] Furthermore, in the small molecule compound searching apparatus 2 of this embodiment, the constraints indicated by the chemical structure constraint information B1 may include a first constraint and a second constraint that applies to more chemical structures of small molecule compounds than the first constraint, and the first chemical structure selection unit 203 may select second small molecule compounds having chemical structures that satisfy the second constraint, and then select second small molecule compounds having chemical structures that satisfy the first constraint from the selected second small molecule compounds. With this configuration, the small molecular weight compound screening apparatus 2 of this embodiment can sequentially select second small molecular weight compounds having chemical structures that satisfy constraints that apply to the chemical structures of many small molecular weight compounds, thereby improving the efficiency of selecting second small molecular weight compounds.

[0087] Moreover, the small molecule compound screening apparatus 2 according to this embodiment may further include a character string conversion section 209 that converts the chemical structure of a small molecule compound into a character string. With this configuration, the small molecule compound searching apparatus 2 of this embodiment can convert a chemical structure into a character string for executing processing according to a program, so that processing can be executed even if the information used for processing does not include the chemical structure as a character string.

[0088] Second Embodiment The second embodiment of the present invention will now be described in detail with reference to the drawings. In the above first embodiment, the case where the chemical structure constraint information B1 is input is described. In the present embodiment, the case where the chemical structure constraint information B1 is generated based on the input target property value is described. The same components as those in the first embodiment described above are denoted by the same reference numerals, and descriptions of the same components and operations may be omitted.

[0089] The low molecular compound screening system according to this embodiment is referred to as low molecular compound screening system 1a, and the low molecular compound screening device is referred to as low molecular compound screening device 2a. The configuration of the low molecular compound screening system 1a is the same as that of the low molecular compound screening system 1 (FIG. 1), except that the low molecular compound screening device 2a is provided instead of the low molecular compound screening device 2.

[0090] [Functional configuration of the low molecular compound searching device 2a] 6 is a diagram showing an example of the functional configuration of a small molecule compound screening apparatus 2a according to this embodiment. The small molecule compound screening apparatus 2a includes a control unit 20a and a storage unit .

[0091] The control unit 20a includes a chemical structure generation unit 200, a structure screening unit 201a, a property screening unit 204, a character string conversion unit 209, and an output unit 210. Comparing the control unit 20a according to this embodiment (FIG. 6) with the control unit 20 according to the first embodiment (FIG. 2), the structure screening unit 201a is different. Here, the functions of the other components (the chemical structure generation unit 200, the property screening unit 204, the character string conversion unit 209, and the output unit 210) are the same as those of the first embodiment.

[0092] The structure screening unit 201a includes a constraint information acquisition unit 202, a first chemical structure selection unit 203, and a constraint condition generation unit 211a. The functions of the first chemical structure selection unit 203 are the same as those in the first embodiment.

[0093] The constraint condition generating unit 211a generates chemical structure constraint information B1 based on the constraint generating information J1. The constraint generating information J1 is information indicating a desired reaction used in the synthesis of a low molecular weight compound or a physical property of the low molecular weight compound. The constraint generating information J1 is obtained by inputting the constraint generating information J1 from the input device 3.

[0094] The constraint information acquisition unit 202 acquires the constraint generation information J1 generated by the constraint condition generation unit 211a.

[0095] [Structure screening process] The process performed by the small molecule compound screening apparatus 2a according to this embodiment is different from the process performed by the small molecule compound screening apparatus 2 according to the first embodiment in the structure screening process. The learning process and the property screening process are the same as those in the first embodiment.

[0096] Fig. 7 is a diagram showing an example of the structure screening process according to this embodiment. Note that the processes in steps S310 and S350 are similar to those in steps S10 and S30 in Fig. 3, and therefore descriptions thereof will be omitted.

[0097] Step S320: The constraint generation information J1 input from the input device 3 is acquired.

[0098] Step S330: The constraint condition generating unit 211a generates the chemical structure constraint information B1 based on the constraint generation information J1. The constraint condition generating unit 211a generates the chemical structure constraint information B1 from the constraint generation information J1 based on, for example, machine learning. When machine learning is used, the constraint condition generating unit 211a generates the chemical structure constraint information B1 from the constraint generation information J1 based on a trained model in which the relationship between the desired reaction used in the synthesis of the low molecular weight compound or the physical properties of the low molecular weight compound and the constraints on the chemical structure required for the compound is learned. As another example, the constraint condition generating unit 211a may generate the chemical structure constraint information B1 from the constraint generation information J1 based on a table indicating the relationship between the desired reaction used in the synthesis of the low molecular weight compound or the physical properties of the low molecular weight compound and the constraints on the chemical structure required for the compound.

[0099] Step S340: The constraint information acquisition unit 202 acquires the constraint generation information J1 generated by the constraint condition generation unit 211a. The constraint information acquisition unit 202 stores the acquired chemical structure constraint information B1 in the storage unit 21. With this, the control unit 20a ends the structure screening process.

[0100] As described above, the small molecule compound searching apparatus 2a according to this embodiment further includes a constraint condition generating unit 211a that generates chemical structure constraint information B1 based on constraint generation information J1 indicating a desired reaction used in the synthesis of a small molecule compound or a physical property of the small molecule compound. With this configuration, the low molecular compound searching apparatus 2a according to this embodiment can search for low molecular compounds simply by inputting the constraint generation information J1 indicating the desired reaction used in the synthesis of the low molecular compound or the physical properties of the low molecular compound. By using the low molecular compound searching apparatus 2a according to this embodiment, even a user who does not have sufficient knowledge of the constraints of the chemical structure corresponding to the desired reaction used in the synthesis of the low molecular compound or the physical properties of the low molecular compound can search for low molecular compounds.

[0101] In addition, a part of the low molecular compound searching apparatus 2, 2a in the above-mentioned embodiment, for example, the chemical structure generating unit 200, the structure screening unit 201, 201a, the physical property screening unit 204, the character string conversion unit 209, and the output unit 210 may be realized by a computer. In that case, a program for realizing this control function may be recorded in a computer-readable recording medium, and the program recorded in the recording medium may be read into a computer system and executed to realize the control function. Note that the "computer system" here refers to a computer system built into the low molecular compound searching apparatus 2, 2a, and includes hardware such as an operating system (OS) and peripheral devices. In addition, the "computer-readable recording medium" refers to a portable medium such as a flexible disk, a magneto-optical disk, a ROM (Read Only Memory), a CD-ROM (Compact Disc-Read Only Memory), and a storage device such as a hard disk built into the computer system. Furthermore, the term "computer-readable recording medium" may include a medium that dynamically stores a program for a short period of time, such as a communication line when transmitting a program via a network such as the Internet or a communication line such as a telephone line, and a medium that stores a program for a certain period of time, such as a volatile memory inside a computer system that serves as a server or client in such a case. The above program may be one that realizes part of the above-mentioned functions, or may be one that can realize the above-mentioned functions in combination with a program already recorded in the computer system. In addition, a part or all of the small molecule compound searching apparatus 2, 2a in the above-mentioned embodiment may be realized as an integrated circuit such as LSI (Large Scale Integration). Each functional block of the small molecule compound searching apparatus 2, 2a may be individually processed, or a part or all of them may be integrated and processed. The integrated circuit method is not limited to LSI, and may be realized by a dedicated circuit or a general-purpose processor. In addition, when an integrated circuit technology that replaces LSI appears due to the progress of semiconductor technology, an integrated circuit based on that technology may be used.

[0102] Although one embodiment of the present invention has been described in detail above with reference to the drawings, the specific configuration is not limited to the above, and various design changes, etc. are possible within the scope that does not deviate from the gist of the present invention. [Explanation of symbols]

[0103] 2, 2a...low molecular weight compound searching device, 202...constraint information acquisition unit, 203...first chemical structure selection unit, 207...regression calculation unit, B1...chemical structure constraint information, F1...trained model (regression model)

Claims

1. a constraint information acquisition unit that acquires chemical structure constraint information indicating constraints on a chemical structure required for a compound; a first chemical structure selection unit that selects, from a plurality of first low molecular weight compounds, a second low molecular weight compound having a chemical structure that satisfies the constraints indicated by the chemical structure constraint information acquired by the constraint information acquisition unit; a regression calculation unit that calculates, for each of the second low molecular weight compounds selected by the first chemical structure selection unit, an estimated value of the physical property value of the second low molecular weight compound from the chemical structure of the second low molecular weight compound based on a regression model generated based on the chemical structure of the low molecular weight compound and the physical property value of the low molecular weight compound; A low molecular weight compound exploration device comprising:

2. The method further includes a second chemical structure selection unit that selects a third low molecular weight compound that satisfies a target physical property value from the second low molecular weight compounds selected by the first chemical structure selection unit based on the estimated value calculated by the regression calculation unit. The low molecular compound screening apparatus according to claim 1 .

3. the regression calculation unit calculates, for each of the second low molecular weight compounds selected by the first chemical structure selection unit, a probability distribution of physical property values ​​of the second low molecular weight compounds from the chemical structures of the second low molecular weight compounds based on the regression model, and calculates an acquisition function from the calculated probability distribution; The low molecular compound screening apparatus according to claim 1 .

4. The regression model is a regression model based on either Gaussian process regression, Bayesian ridge regression, ensemble regression, deep learning, or gradient boosting based regression. The low molecular weight compound screening device according to claim 3 .

5. A chemical structure generating unit is further provided which generates a structural formula showing a chemical structure of the first low molecular weight compound for each of the plurality of first low molecular weight compounds. The low molecular compound screening apparatus according to claim 1 .

6. The first plurality of small molecule compounds include at least one of compounds selected by a human, compounds extracted from a database, and compounds designed by a specific algorithm. The low molecular compound screening apparatus according to claim 1 .

7. The physical property values ​​used in generating the regression model are at least one of the actual measured values ​​of the physical property values ​​of the low molecular weight compound, or calculated or predicted values ​​calculated by a specific algorithm. The low molecular compound screening apparatus according to claim 1 .

8. a constraint condition generating unit that generates the chemical structure constraint information based on constraint generation information indicating a desired reaction used in the synthesis of a low molecular weight compound or a physical property of the low molecular weight compound; The low molecular compound screening apparatus according to claim 1 .

9. the constraints indicated by the chemical structure constraint information include a first constraint and a second constraint that applies to more chemical structures of low molecular weight compounds than the first constraint; The first chemical structure selection unit selects second low molecular weight compounds having a chemical structure that satisfies the second constraint, and then selects second low molecular weight compounds having a chemical structure that satisfies the first constraint from the selected second low molecular weight compounds. The low molecular compound screening apparatus according to claim 1 .

10. Further comprising a character string conversion unit for converting the chemical structure of the low molecular weight compound into a character string; The low molecular compound screening apparatus according to claim 1 .

11. obtaining chemical structure constraint information indicating constraints on a chemical structure required for the compound; selecting a second low molecular weight compound having a chemical structure that satisfies the constraint indicated by the acquired chemical structure constraint information from among the plurality of first low molecular weight compounds; calculating, for each of the selected second low molecular weight compounds, an estimated value of the physical property value of the second low molecular weight compound from the chemical structure of the second low molecular weight compound based on a regression model generated based on the chemical structure of the low molecular weight compound and the physical property value of the low molecular weight compound; Low molecular compound search method.

12. On the computer, obtaining chemical structure constraint information indicating constraints on a chemical structure required for the compound; selecting a second low molecular weight compound having a chemical structure that satisfies the constraint indicated by the acquired chemical structure constraint information from among the plurality of first low molecular weight compounds; calculating, for each of the selected second low molecular weight compounds, an estimated value of the physical property value of the second low molecular weight compound from the chemical structure of the second low molecular weight compound based on a regression model generated based on the chemical structure of the low molecular weight compound and the physical property value of the low molecular weight compound; A program for executing the above.

Images