JP7347147B2 - Molecular descriptor generation system, molecular descriptor generation method, and molecular descriptor generation program - Google Patents

Description

One aspect of the present disclosure relates to a molecular descriptor generation system, a molecular descriptor generation method, and a molecular descriptor generation program.

Conventionally, the structure of a molecule is obtained in a predetermined format, converted into vector information, and the converted information is input into a machine learning algorithm or the like to be used for property prediction or the like. For example, a method is known that uses machine learning to predict the bond between the three-dimensional structure of a biopolymer and the three-dimensional structure of a compound (see Patent Document 1 below). In this method, a predicted 3D structure of a complex between a biopolymer and a compound is generated based on the 3D structure of the biopolymer and the 3D structure of the compound, and the predicted 3D structure is converted into a predicted 3D structure vector. Using a learning algorithm, the predicted 3D structure vectors are used as data for predicting the bond between the 3D structure of a biopolymer and the 3D structure of a compound.

JP 2019-28879 Publication

In recent years, there has been a demand for data representation of the structures of compounds synthesized from molecules of multiple types of constituent units in various composition ratios. However, in the conventional technique described in Patent Document 1, if there is little accumulation of data on the three-dimensional structure of the compound, it is difficult to generate data that appropriately represents the molecular structure of the compound. Therefore, even if the molecular structure of a compound synthesized from multiple types of structural units is not clear, a mechanism is desired that can appropriately represent the molecular structure of the compound as data.

A molecular descriptor generation system according to an embodiment of the present disclosure includes at least one processor, and the at least one processor is configured to generate a molecular descriptor for each molecular structure of a plurality of types of constituent unit molecules and a combination of the molecular structure with an external molecular structure. Accepts at least input of structural data representing a point and composition ratio data representing the composition ratio of molecules of multiple types of constitutional units, and includes molecules of multiple types of constitutional units indicated by the structural data at the composition ratio indicated by the composition ratio data. Data for a cyclic structure with a predetermined number of structural units is repeatedly generated by randomly arranging molecules of multiple types of structural units in a ring, and then bonding adjacent molecules at bonding points indicated by the structural data. By converting the repeatedly generated data of multiple cyclic structures into molecular descriptions, converting them into multiple vectors, and combining the multiple vectors, we can create molecules of synthetic compounds containing molecules of multiple types of constituent units in the composition ratio. Generate written molecular description data.

Alternatively, a molecular descriptor generation method according to another aspect of the present disclosure is a molecular descriptor generation method executed by a computer including at least one processor, which comprises a step of receiving at least input of structural data representing bonding points with external molecular structures in the molecular structure and composition ratio data representing the composition ratio of the molecules of the plurality of types of structural units; and a step of receiving inputs of the plurality of types of structural units indicated by the structural data. The data of a cyclic structure with a predetermined number of structural units containing molecules in the composition ratio indicated by the composition ratio data are obtained by randomly arranging molecules of multiple types of constituent units in a ring, and then dividing the adjacent molecules between adjacent molecules according to the structural data. A step of repeatedly generating multiple cyclic structures by combining them at the bonding points shown, a step of converting the data of the repeatedly generated multiple cyclic structures into multiple vectors by converting them into molecular descriptions, and a step of converting the multiple vectors into multiple vectors by combining the multiple vectors. The method includes the step of generating molecule description data describing molecules of a synthetic compound containing molecules of different types of constituent units in a composition ratio.

Alternatively, a molecular descriptor generation program according to another embodiment of the present disclosure may provide a computer with structural data representing the molecular structures of molecules of multiple types of constituent units and bonding points with external molecular structures in the molecular structures; a step of receiving at least an input of composition ratio data representing the composition ratio of molecules of the plurality of types of structural units; and a predetermined number of constituent units containing molecules of the plurality of types of composition units indicated by the structural data at the composition ratio indicated by the composition ratio data. The data of a cyclic structure is repeatedly generated by randomly arranging molecules of multiple types of constituent units in a ring, and then bonding adjacent molecules at bonding points indicated by the structural data. By converting the data of multiple cyclic structures into molecular descriptions into multiple vectors, and by combining the multiple vectors, we describe the molecules of synthetic compounds containing molecules of multiple types of constituent units in the composition ratio. and generating molecular description data.

According to the aspect of the present disclosure, the molecular structure of a synthetic compound in which multiple types of structural units are synthesized can be appropriately expressed as data.

1 is a diagram illustrating an example of a hardware configuration of a computer that constitutes a molecular descriptor generation system according to an embodiment. FIG. 1 is a diagram illustrating an example of a functional configuration of a molecular descriptor generation system according to an embodiment. 3 is a diagram illustrating an image of a molecular structure indicated by structural data acquired by the acquisition unit 11 of FIG. 2. FIG. FIG. 3 is a diagram illustrating a specific example of a molecular structure indicated by structural data acquired by the acquisition unit 11 in FIG. 2 using a structural formula. FIG. 3 is a diagram illustrating a specific example of a molecular structure indicated by structural data acquired by the acquisition unit 11 in FIG. 2 using a structural formula. 3 is a diagram showing an image of a molecular structure specified by the cyclic structure data SDr generated by the data generation unit 12 of FIG. 2. FIG. 3 is a diagram showing an image of a molecular structure specified by linear structure data SDs generated by the data generation unit 12 of FIG. 2. FIG. 1 is a flowchart illustrating an example of the operation of the molecular descriptor generation system according to the embodiment. FIG. 3 is a diagram showing a numerical example of a vector generated by the molecular descriptor generation system according to the embodiment. FIG. 3 is a diagram showing a numerical example of a vector generated by the molecular descriptor generation system according to the embodiment. FIG. 3 is a diagram showing a numerical example of a vector generated by the molecular descriptor generation system according to the embodiment.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. In the description, the same elements or elements having the same function will be denoted by the same reference numerals, and redundant description will be omitted.

[System overview]
The molecular descriptor generation system 10 according to the embodiment generates molecular descriptors ( This is a computer system that executes the generation process of molecular description data. A constituent unit molecule refers to a molecule that constitutes a material used to synthesize a synthetic compound, and is, for example, a monomer that is a single polymer. A synthetic compound is a chemical substance that is produced by synthesizing multiple types of constituent unit molecules in a predetermined composition ratio. For example, if the constituent unit molecules are monomers, they are formed by linking multiple types of monomers. It is a copolymer. Preferably, the synthetic compound to be subjected to the molecular descriptor generation process of the molecular descriptor generation system 10 of the present embodiment has a linear structure in which a plurality of types of constituent unit molecules are connected in a linear chain. It is a compound of

Input data generated by the molecular descriptor generation system 10 is used as input data for machine learning to predict properties of synthetic compounds. The properties of the synthetic compound include, for example, glass transition temperature, elastic modulus, dielectric constant, dielectric loss tangent, thermal expansion coefficient, and the like. Machine learning, in which input data is input, is a method of autonomously finding laws or rules by iteratively learning based on given information. The specific method of machine learning is not limited. For example, machine learning may be machine learning using a machine learning model that is a calculation model including a neural network. A neural network is an information processing model that mimics the structure of the human nervous system. As more specific examples, machine learning includes graph neural networks (GNN), convolutional neural networks (CNN), recurrent neural networks (RNN), attention RNNs, and multi-head attention (Multi-Head Attention). ), random forest, support vector machine, and multiple regression.

[System configuration]
The molecular descriptor generation system 10 is composed of one or more computers. When a plurality of computers are used, one logical molecular descriptor generation system 10 is constructed by connecting these computers via a communication network such as the Internet or an intranet.

FIG. 1 is a diagram showing an example of a general hardware configuration of a computer 100 that constitutes a molecular descriptor generation system 10. For example, the computer 100 includes a processor (e.g., CPU) 101 that executes an operating system, application programs, etc., a main storage section 102 made up of ROM and RAM, and an auxiliary storage section 103 made up of a hard disk, flash memory, etc. , a communication control unit 104 configured with a network card or a wireless communication module, an input device 105 such as a keyboard, a mouse, a touch panel, etc., and an output device 106 such as a monitor, a touch panel display, etc.

Each functional element of the molecular descriptor generation system 10 is realized by loading a predetermined program into the processor 101 or the main storage unit 102 and causing the processor 101 to execute the program. The processor 101 operates the communication control unit 104, the input device 105, or the output device 106 according to the program, and reads and writes data in the main storage unit 102 or the auxiliary storage unit 103. Data or databases required for processing are stored in the main storage unit 102 or the auxiliary storage unit 103.

FIG. 2 is a diagram showing an example of the functional configuration of the molecular descriptor generation system 10. The molecular descriptor generation system 10 includes an acquisition section 11, a data generation section 12, a vector conversion section 13, a vector synthesis section 14, and a weighted addition section 15 as functional elements.

The acquisition unit 11 obtains structural data of a plurality of types of constituent unit molecules, and composition ratios of each of the plurality of constituent unit molecules when it is assumed that a synthetic compound is generated by synthesizing these plurality of constituent unit molecules. This is a functional element that accepts input of composition ratio data representing . The acquisition unit 11 may acquire these data from a database within the molecular descriptor generation system 10 in response to selection input by the user of the molecular descriptor generation system 10, or may acquire these data from an external computer or the like in response to selection by the user. You may also obtain it by

Specifically, the acquisition unit 11 acquires structural data specifying the molecular structures of multiple types of constituent unit molecules in a specific data format. This specific data format is not limited to any specific format as long as it is a format that can express the molecular structure, but it may be an image format such as JPEG or GIF that images the molecular structure, or it may be a character string that represents the molecular structure. , coordinates, etc., or a molecular graph data format that specifies the structure of an undirected graph in which the molecular structure is expressed by nodes and edges using numbers, alphabets, text, vectors, etc. The data may be data in any combination of two or more of these data formats. It may also be a one-dimensional character notation that represents the molecular structure as an array of characters, generally referred to as SMILES, InChl, SLN, etc. The individual numbers that make up this particular data format may be expressed in decimal notation, or in other notations such as binary or hexadecimal. Here, the acquisition unit 11 acquires the structural data including data regarding bonding points that can be chemically bonded to an external molecular structure in the molecular structure of the constituent unit molecule.

FIG. 3 shows an image of the molecular structure indicated by the structural data obtained by the obtaining section 11, and FIGS. 4 and 5 show specific examples of the molecular structure indicated by the structural data obtained by the obtaining section 11 using structural formulas. ing. As shown in FIG. 3, the acquisition unit 11 includes, for example, data representing the molecular structure "A" and data representing the position of the bonding point in the molecular structure "A" (for example, data representing the position with "*"). ) is obtained, and similar structural data SD _{2 ,} SD ₃ , and SD ₄ are obtained regarding molecular structures “B”, “C”, and “D” of a plurality of types of constituent unit molecules. The molecular structures of the structural data SD ₅ and SD ₆ shown in FIG. 4 are examples in which the constituent unit molecules are propylene and butylene. In this way, when there are two types of constituent unit molecules, propylene and butylene, a polymer compound (copolymer) can be produced by addition polymerization of multiple constituent unit molecules with each other. The position "*" of an equivalent bonding point where the bonding destination is not limited is indicated. The molecular structures of structural data SD ₇ and SD ₈ shown in FIG. 5 are examples of terephthalic acid and ethylene glycol. In this example, a polymer compound (copolymer) can be produced by condensation polymerization of multiple constituent unit molecules, so unequal bonding points with limited bonding destinations are formed at both ends of each two types of molecules. Positions “* ₁ ” and “* ₂ ” are shown. These indications “* ₁ ” and “* ₂ ” indicate that bonding points with the same identification number will not bond with each other, but bonding points with different identification numbers may bond with each other. It will be done.

The acquisition unit 11 also acquires composition ratio data regarding the constituent unit molecules of the molecular structure specified by the plurality of types of structural data described above. At this time, the acquisition unit 11 may acquire data indicating the composition ratio of each constituent unit molecule itself as the mixture ratio data representing the composition ratio ra of the plurality of types of constituent unit molecules, or may acquire data indicating the composition ratio of each constituent unit molecule itself. Data indicating the composition ratio between the two may also be obtained. For example, when the structural data of the molecular structure shown in FIG. 3 is obtained, the composition ratio ra ₁ = "0.2" of the molecular structure "A" and the composition ratio ra ₂ = "0.2" of the molecular structure "B". 3'', the composition ratio ra ₃ = "0.1" of the molecular structure "C", and the composition ratio ra ₄ = "0.4" of the molecular structure "D".

Further, the acquisition unit 11 generates a combination of structural data of the plurality of types of constituent unit molecules and composition ratio data representing the composition ratio of the plurality of types of constituent unit molecules, which are synthesized from the plurality of types of constituent unit molecules. A plurality of pieces of data are acquired for each synthetic compound, and mixture rate data representing a mixture rate of a plurality of types of synthetic compounds corresponding to the combination thereof is further acquired. Thereby, by using the functions described below, it is possible to generate molecular description data that describes the molecular structure of a mixture obtained by mixing a plurality of types of synthetic compounds at a mixing ratio indicated by the mixing ratio data. At this time, the obtaining unit 11 may obtain data indicating the mixing ratio of each synthetic compound itself as the mixing ratio data representing the mixing ratio rb of the plurality of synthetic compounds, or may obtain data indicating the mixing ratio of the plurality of synthetic compounds. Data indicating the ratio may also be obtained.

The data generation unit 12 refers to the combinations of structural data and composition ratio data and generates structural data that specifies the molecular structure of the synthetic compound corresponding to each combination. That is, the data generation unit 12 generates circular structure data and linear structure data as structural data.

Specifically, the data generation unit 12 constructs structural data of a cyclic structure having a predetermined number of structural units that includes a plurality of types of structural unit molecules indicated by the structural data at a composition ratio indicated by the composition ratio data. For example, regarding the constituent unit molecules "A", "B", "C", and "D" as shown in FIG. 3, their composition ratios are "0.2", "0.3", "0.1", When "0.4" is specified, the four types of constituent unit molecules "A", "B", "C", and "D" have the specified composition ratios of "0.2" and "0. .3", "0.1", and "0.4", and are arranged in a ring so that the total number of constituent units is a predetermined number (for example, 200). . The number of constituent units needs to be at least the minimum number that can express the ratio. For example, when the composition ratio of the constituent unit molecule "A" is 0.1 and the composition ratio of the constituent unit molecule "B" is 0.001, the number of constituent units is 101 or more. Then, in the cyclic structure, structural data is generated in which the orientation of each constituent unit molecule is adjusted so that adjacent constituent unit molecules are bonded at the bonding points indicated by the structural data. . At this time, if the structural data indicates that the constituent unit molecules have unequal bonding points, the data generation unit 12 assigns different identification numbers to the bonding points in adjacent constituent unit molecules. The molecular arrangement or molecular orientation in the cyclic structure is finely adjusted so that it binds to the bonding point in the cyclic structure.

Further, the data generation unit 12 constructs structural data of a linear structure having a predetermined number of structural units including a plurality of types of structural unit molecules indicated by the structural data at a composition ratio indicated by the composition ratio data. For example, regarding the constituent unit molecules "A", "B", "C", and "D" as shown in FIG. 3, their composition ratios are "0.2", "0.3", "0.1", When "0.4" is specified, the four types of constituent unit molecules "A", "B", "C", and "D" have the specified composition ratios of "0.2" and "0. .3", "0.1", and "0.4", and the units are randomly selected to form a predetermined number of constituent units (for example, 20,000 units) in series (in series). Arranged in a chain). This predetermined number of structural units is set in advance so that the concentration of molecular terminals in the linear structure (ratio of unbonded bonding points to the total number of structural units) is below a predetermined value (for example, 0.01%). has been done. Then, in the linear structure, structural data is generated in which the orientation of each constituent unit molecule is adjusted so that adjacent constituent unit molecules are bonded at the bonding points indicated by the structural data. Ru. At this time, if the structural data indicates that the constituent unit molecules have unequal bonding points, the data generation unit 12 assigns different identification numbers to the bonding points in adjacent constituent unit molecules. The molecular arrangement or molecular orientation in the linear structure is finely adjusted so that it binds to the bonding point in the linear structure.

Then, the data generation unit 12 repeatedly generates the cyclic structure data and the linear structure data regarding the synthetic compound corresponding to the combination of the structural data and the composition ratio data a predetermined number of times (for example, 100 times each) in a plurality of ways. cyclic structure data and multiple linear structure data are generated. FIG. 6 shows the composition ratios of "0.2", "0.3", "0. 1" and "0.4" are specified, and FIG. An image of the molecular structure specified by data SDs is shown. Furthermore, the data generation unit 12 generates multiple types of cyclic structure data and multiple types of linear structure data for each synthetic compound corresponding to the combination of the structural data and composition ratio data acquired by the acquisition unit 11. repeat.

The vector conversion unit 13 converts the plurality of types of cyclic structure data and linear structure data for each synthetic compound generated by the data generation unit 12 into one-dimensional vectors Vr and Vs by respectively converting them into molecular descriptions. (Figure 6, Figure 7). Molecular description allows the characteristics of a molecule indicated by structural data to be expressed as a numerical sequence based on its chemical structure. Any method that vectorizes the molecular structure can be adopted as this molecular description method, but examples include ECFP (Extended Connectivity FingerPrints), MACCS FingerPrints, PubChem FingerPrints, Substructure FingerPrints, Estate FingerPrints, and BCI FingerPrints. , Molprint2D FingerPrints, Pass base FingerPrints, etc. can be adopted.

The vector synthesis unit 14 combines a plurality of vectors Vr converted from a plurality of cyclic structure data regarding a synthetic compound corresponding to the combination of structure data and composition ratio data, thereby describing the molecule of the synthetic compound in a numerical string. molecule description data MDr is generated. That is, the vector synthesis unit 14 adds each element of a plurality of vectors Vr generated corresponding to the synthetic compound, and then converts each added element into the total molecular weight of the plurality of cyclic structures or the total component molecule. By dividing by a number, molecular description data MDr corresponding to the synthetic compound is generated. For example, two types of component molecules with molecular weights of 42.08 and 56.11 are arranged in a 1:1 composition ratio to generate data for a cyclic structure with a total of 200 components, and this is repeated 100 times. In this case, each added element is divided by the total molecular weight = 420,800 + 561,100 = 981,900, or the total number of elements = 20,000. Similarly, by combining a plurality of vectors Vs converted from a plurality of linear structure data regarding a synthetic compound, molecular description data MDs that describes the molecule of the synthetic compound in a numerical string is generated. That is, the vector synthesis unit 14 adds each element of the plurality of vectors Vs generated corresponding to the synthetic compound, and then converts each added element into the total molecular weight or the total component of the plurality of linear structures. By dividing by the number of molecules, molecular description data MDs corresponding to the synthetic compound is generated.

Then, the vector synthesis unit 14 repeats the creation of the two molecular description data MDr and MDs for each synthetic compound corresponding to the combination of the structural data and the composition ratio data acquired by the acquisition unit 11.

The weighted addition unit 15 weights and adds the molecular description data MDr for each of the plurality of types of synthetic compounds generated by the vector synthesis unit 14, thereby creating a mixture in which the plurality of types of synthetic compounds are mixed at a mixing ratio indicated by the mixing ratio data. Final molecular description data MDr' indicating the characteristics of the chemical structure of the material is generated. That is, the weighted addition unit 15 weights and adds each element of the vector of a plurality of molecular description data MDr by the mixing ratio of the synthetic compound described by each molecular description data MDr, thereby obtaining the molecular description data MDr'. generate. Similarly, by weighting and adding the molecular description data MDs for each of the plurality of synthetic compounds generated by the vector synthesis unit 14, a mixed material obtained by mixing the plurality of synthetic compounds at the mixing ratio indicated by the mixing ratio data is obtained. Final molecular description data MDs' indicating the characteristics of the chemical structure is generated. That is, the weighted addition unit 15 weights and adds each element of the vector of a plurality of molecule description data MDs by the mixing ratio of the synthetic compound described by each molecule description data MDs, thereby obtaining the molecule description data MDs'. generate.

Furthermore, the weighted addition unit 15 outputs the generated molecular description data MDr', MDs' to the outside. The output molecular description data MDr', MDs' are read as input data by a training unit 20 in a computer connected to the outside of the molecular descriptor generation system 10. Then, in the training unit 20, the input data is input to the machine learning model as an explanatory variable together with an arbitrary teacher label, thereby generating a learned model. Furthermore, a machine learning model within the predictor 30 is set based on the learned model generated by the training unit 20. The molecular description data MDr' and MDs' generated by the molecular descriptor generation system 10 are input to the machine learning model in the predictor 30, so that the predictor 30 generates and generates prediction results of the properties of the mixed material. Output. Note that the training unit 20 and the predictor 30 may be configured in the same computer as the computer 100 that configures the molecular descriptor generation system 10, or may be configured in a separate computer from the computer 100. good.

In one example, the machine learning model generated by the training unit 20 is a trained model that is expected to have the highest estimation accuracy, and therefore can be called the "best machine learning model." However, it should be noted that this trained model is not necessarily "actually the best". A trained model is generated by a computer processing training data that includes many combinations of input data and output data. The computer calculates output data by inputting input data into a machine learning model, and calculates the error between the calculated output data and the output data indicated by the teacher data (i.e., the difference between the estimation result and the correct answer). . The computer then updates a given parameter in the machine learning model based on that error. A computer generates a trained model by repeating such learning. The process of generating a learned model can be called a learning phase, and the process of the predictor 30 that uses the learned model can be called an operation phase.

[System operation]
With reference to FIG. 8, the operation of the molecular descriptor generation system 10 will be explained, and the molecular descriptor generation method according to this embodiment will be explained. FIG. 8 is a flowchart showing an example of the operation of the molecular descriptor generation system 10, and FIGS. 9, 10, and 11 are diagrams showing numerical examples of each vector generated by the molecular descriptor generation system 10.

First, when molecular descriptor generation processing is started in response to an instruction input by a user of the molecular descriptor generation system 10, the acquisition unit 11 collects a plurality of structural data and composition ratio data for each of a plurality of types of synthetic compounds. Mixing ratio data regarding combinations and mixing ratios of multiple types of synthetic compounds is acquired (step S1). Thereafter, the data generation unit 12 generates a plurality of types of cyclic structure data and a plurality of types of linear structure data for the combination of the above-mentioned structure data and composition ratio data (step S2).

Next, the vector conversion unit 13 converts the plurality of types of cyclic structure data and the plurality of types of linear structure data into vectors Vr and Vs by molecular description, respectively (step S3). 9 and 10 show numerical examples of vectors Vr and Vs generated for the combination of the structural data specifying the two molecular structures shown in FIG. 4 and the composition ratio data indicating the composition ratio of 1:1.

Furthermore, the vector synthesis unit 14 combines the plurality of vectors Vr converted from the plurality of types of cyclic structure data to generate molecular description data MDr, and generates the molecular description data MDr by combining the plurality of vectors Vr converted from the plurality of types of linear structure data. Vs are combined to generate molecular description data MDs (step S4). FIG. 11 shows a numerical example of the molecular description data MDr, which is a one-dimensional vector generated by the vector synthesis unit 14.

Thereafter, the processes of steps S2 to S4 described above are repeated for the combination of the structural data and the composition ratio data acquired by the acquisition unit 11, and as a result, the molecular description data MDr and the molecular description A plurality of combinations of data MDs are generated (step S5). Then, the weighted addition unit 15 weights and adds each of the molecular description data MDr and MDs corresponding to a plurality of types of synthetic compounds at the mixing ratio indicated by the mixing ratio data, thereby obtaining final molecular description data MDr', MDs' are generated as input data (step S6).

Next, in the training unit 20, a learning phase is executed, and a learned model is generated by repeating training using input data and teacher data (step S7). Then, the generated trained model is set in the predictor 30, and the predictor 30 executes the operation phase using input data newly acquired from the molecular descriptor generation system 10 to predict the properties of the mixed material. Results are generated and output (step S8).

[program]
A molecular descriptor generation program for causing a computer or a computer system to function as the molecular descriptor generation system 10 includes an acquisition unit 11, a data generation unit 12, a vector conversion unit 13, a vector synthesis unit 14, and a weighted addition unit. It includes a program code for functioning as 15. This molecular descriptor generation program may be provided after being permanently recorded on a tangible recording medium such as a CD-ROM, DVD-ROM, or semiconductor memory. Alternatively, the molecular descriptor generation program may be provided via a communication network as a data signal superimposed on a carrier wave. The provided molecular descriptor generation program is stored in the auxiliary storage unit 103, for example. The processor 101 reads the molecular descriptor generation program from the auxiliary storage unit 103 and executes it, thereby realizing each of the above functional elements.

[effect]
As explained above, according to the above embodiment, data of a cyclic structure having a predetermined number of structural units including a plurality of types of structural unit molecules in a prespecified composition ratio are obtained by randomly arranging the structural unit molecules. is generated as structural data in which adjacent molecules are bonded at prespecified bonding points. Such data represents the structure of a synthetic compound containing multiple types of constituent unit molecules in their composition ratios. Then, the repeatedly generated data of the plurality of cyclic structures is converted into a plurality of vectors, and the plurality of vectors are combined to generate molecular description data that appropriately expresses the molecular structure of the synthetic compound. As a result, even if the molecular structure of a compound synthesized from multiple types of structural units is not clear, the molecular structure of the compound can be appropriately expressed as data. In particular, with such a function, it is possible to efficiently generate molecular structure data with a small proportion of reactive terminals using a small number of structural units, and therefore it is also possible to reduce the processing load when generating molecular description data. Further, by using data on a cyclic structure, it is possible to generate data that more appropriately represents the molecular structure of a synthetic compound having a linear structure. In addition, when the constituent unit molecules are set as monomers, it is possible to generate data that more appropriately represents the molecular structure of the copolymer, which is a synthetic compound.

Furthermore, in the above embodiment, when generating the molecular description data MDr, after adding the elements of a plurality of vectors Vr, the elements are expressed as the total molecular weight or the total number of constituent unit molecules in the plurality of cyclic structures. It is dividing. In this way, data expressing the molecule of one synthetic compound can be generated using the molecular structures of a plurality of randomly generated cyclic structures without being influenced by the size of the molecular structures.

Further, according to the above embodiment, data of a linear structure having a predetermined number of constituent units including multiple types of constituent unit molecules at a prespecified composition ratio are arranged in such a manner that the constituent unit molecules are randomly arranged and then adjacent to each other. It is generated as data of a structure in which molecules are bonded at prespecified bonding points. Such data also represent the structure of a synthetic compound containing multiple types of constituent unit molecules in their composition ratios. Then, the data of multiple linear structures that are repeatedly generated are converted into multiple vectors Vs, and these multiple vectors Vs are further combined to create molecular description data MDs that appropriately expresses the molecular structure of the synthetic compound. is generated. Thereby, even if the molecular structure of a synthetic compound is not clear, the molecular structure of the synthetic compound can be appropriately expressed as data. In particular, when the structure of a synthetic compound is often a linear structure, the molecular structure of the synthetic compound can be expressed more appropriately with data.

In addition, in the above embodiment, when generating the molecular description data MDs, after adding the elements of a plurality of vectors Vs, the elements are calculated as It is divided by In such a configuration, the molecular structures of multiple randomly generated linear structures are used to generate data representing the molecule of one synthetic compound without being influenced by the size of those molecular structures. be able to.

Further, in the above embodiment, molecular description data MDr', MDs', which is a vector obtained by weighting the molecular description data MDr, MDs generated for multiple types of synthetic compounds by the mixture ratio indicated by the mixture ratio data, is further generated. There is. If such a configuration is adopted, it is possible to appropriately generate a vector representing a molecule of a mixture of a plurality of types of synthetic compounds.

[Modified example]
The present invention has been described above in detail based on the embodiments thereof. However, the present invention is not limited to the above embodiments. The present invention can be modified in various ways without departing from the gist thereof.

In the above embodiment, the molecular description data generated by the molecular descriptor generation system 10 is used as input data for machine learning, but the molecular description data may be used for other purposes. For example, it may be used for search processing to search for structures, physical properties, etc. of synthetic compounds or mixtures thereof.

Further, in the above embodiment, the molecular description data MDr', MDs', which is a vector weighted by the mixture ratio indicated by the mixture ratio data, is generated and outputted as the final data. The data MDr and MDs may be used as the final output. Alternatively, a vector combining the two molecular description data MDr and MDs may be used as the final output.

The processing procedure of the molecular descriptor generation method executed by at least one processor is not limited to the example in the above embodiment. For example, some of the steps (processes) described above may be omitted, or each step may be executed in a different order. Furthermore, any two or more of the steps described above may be combined, or some of the steps may be modified or deleted. Alternatively, other steps may be performed in addition to each of the above steps. For example, the processes in steps S7 and S8 may be omitted.

In this disclosure, the expression "at least one processor executes a first process, executes a second process, ... executes an n-th process" or an expression corresponding to this The concept includes a case where the executing entity (that is, the processor) of n processes from the first process to the nth process changes midway. That is, this expression indicates a concept that includes both a case in which all of the n processes are executed by the same processor, and a case in which the processors in the n processes are changed according to an arbitrary policy.

DESCRIPTION OF SYMBOLS 10... Molecular descriptor generation system, 100... Computer, 101... Processor, 11... Acquisition unit, 12... Data generation unit, 13... Vector conversion unit, 14... Vector synthesis unit, 15... Weighted addition unit, 20... Training unit, 30...Predictor, _SD1 to _SD8 ...structural data, SDr...cyclic structure data, SDs...linear structure data, Vr, Vs...vector.

Claims (9)

comprising at least one processor;
the at least one processor,
Inputting structural data representing the molecular structures of molecules of multiple types of structural units and bonding points with external molecular structures in the molecular structures, and composition ratio data representing the composition ratios of the molecules of the multiple types of structural units. At least accept
Data of a cyclic structure having a predetermined number of structural units containing molecules of the plurality of types of structural units indicated by the structural data at the composition ratio indicated by the composition ratio data are obtained by randomly forming molecules of the plurality of types of constituent units into a ring shape. After arranging, repeatedly generate by bonding adjacent molecules at the bonding point indicated by the structural data,
By converting the repeatedly generated data of the plurality of cyclic structures into a plurality of vectors by molecular description,
generating molecular description data describing molecules of a synthetic compound containing molecules of the plurality of types of structural units in the composition ratio by combining the plurality of vectors;
Molecular descriptor generation system. The synthetic compound is a compound with a linear structure,
The molecular descriptor generation system according to claim 1. The molecules of the plurality of types of structural units are monomers,
The synthetic compound is a copolymer,
The molecular descriptor generation system according to claim 1 or 2. The at least one processor includes:
After adding the elements of the plurality of vectors, the element is divided by the total molecular weight in the plurality of cyclic structures or the total number of molecules of the structural unit, thereby generating the molecule description data.
The molecular descriptor generation system according to any one of claims 1 to 3. The at least one processor includes:
The data of a linear structure having a predetermined number of structural units containing molecules of the plurality of types of structural units indicated by the structural data at the composition ratio indicated by the composition ratio data are obtained by randomly arranging the molecules of the plurality of constituent units in series. and then repeatedly generate by bonding adjacent molecules at the bonding points indicated by the structural data,
Converting data of the plurality of linear structures repeatedly generated into a plurality of additional vectors by molecular description,
generating the molecular description data by further combining the plurality of additional vectors;
The molecular descriptor generation system according to any one of claims 1 to 4. The at least one processor includes:
generating the molecular description data by adding the elements of the plurality of additional vectors and then dividing the element by the total molecular weight in the plurality of linear structures or the total number of molecules of the building blocks; ,
The molecular descriptor generation system according to claim 5. The at least one processor includes:
We further accept mixture ratio data representing the mixture ratio of multiple types of synthetic compounds,
further generating a vector in which the molecular description data generated for the plurality of types of synthetic compounds is weighted by a mixing ratio indicated by the mixing ratio data;
The molecular descriptor generation system according to any one of claims 1 to 6. 1. A method for generating molecular descriptors performed by a computer comprising at least one processor, the method comprising:
Inputting structural data representing the molecular structures of molecules of multiple types of structural units and bonding points with external molecular structures in the molecular structures, and composition ratio data representing the composition ratios of the molecules of the multiple types of structural units. a step of accepting at least
Data of a cyclic structure having a predetermined number of structural units containing molecules of the plurality of types of structural units indicated by the structural data at the composition ratio indicated by the composition ratio data are obtained by randomly forming molecules of the plurality of types of constituent units into a ring shape. After arranging, repeatedly generating by bonding adjacent molecules at the bonding points indicated by the structural data;
converting the data of the plurality of repeatedly generated annular structures into a plurality of vectors by molecular description;
generating molecular description data describing molecules of a synthetic compound containing molecules of the plurality of types of structural units in the composition ratio by combining the plurality of vectors;
A molecular descriptor generation method comprising: to the computer,
Inputting structural data representing the molecular structures of molecules of multiple types of structural units and bonding points with external molecular structures in the molecular structures, and composition ratio data representing the composition ratios of the molecules of the multiple types of structural units. a step of accepting at least
Data of a cyclic structure having a predetermined number of structural units containing molecules of the plurality of types of structural units indicated by the structural data at the composition ratio indicated by the composition ratio data are obtained by randomly forming molecules of the plurality of types of constituent units into a ring shape. After arranging, repeatedly generating by bonding adjacent molecules at the bonding points indicated by the structural data;
converting the data of the plurality of repeatedly generated annular structures into a plurality of vectors by molecular description;
generating molecular description data describing molecules of a synthetic compound containing molecules of the plurality of types of structural units in the composition ratio by combining the plurality of vectors;
A molecular descriptor generation program that runs

Images