JP7448353B2 - Computer operating methods, programs, information processing devices, and data structures - Google Patents

Description

The present disclosure relates to a computer operating method, program, information processing device, and data structure.

When using a computer to predict properties such as the refractive index, elastic modulus, or glass transition temperature of a polymer based on information about the molecular structure of the polymer, the molecular structure of the polymer is generally a molecule that can be identified by the computer. It is converted into a descriptor such as a descriptor. For example, Non-Patent Document 1 discloses a method for predicting the refractive index of a polymer by using a number density descriptor and machine learning.

Takuya Minami, Yoshishige Okuno, "Number Density Descriptor on Extended-Connectivity Fingerprints Combined with MachineLearning Approaches for Predicting Polymer Properties", MRS Advances, 3(49), 2975-2980, 2018

In the prior art, there is room for improvement in terms of improving the prediction accuracy of polymer properties.

Therefore, an object of the present disclosure is to provide a computer operating method etc. that can improve the prediction accuracy of polymer properties.

A computer operating method according to one aspect of the present disclosure includes:
The computer is
determining repeating units of the polymer based on information about the molecular structure of the polymer;
generating a molecular descriptor corresponding to a cyclic structure formed by cyclizing the repeating unit;
including.

A program according to one aspect of the present disclosure is
to the computer,
determining repeating units of the polymer based on information about the molecular structure of the polymer;
generating a molecular descriptor corresponding to a cyclic structure formed by cyclizing the repeating unit;
Execute.

An information processing device according to one aspect of the present disclosure includes:
determining the repeating unit of the polymer based on information about the molecular structure of the polymer;
A molecular descriptor corresponding to a cyclic structure formed by cyclizing the repeating unit is generated.

A data structure according to one aspect of the present disclosure is:
A data structure used as training data for an information processing device to generate a learning model for predicting properties of a polymer by machine learning, the molecular description corresponding to a cyclic structure formed by cyclizing repeating units of the polymer. Including children.

According to embodiments of the present disclosure, it is possible to improve the prediction accuracy of polymer properties.

FIG. 1 is a functional block diagram of an information processing device according to an embodiment. 3 is a flowchart illustrating a first processing example by the information processing device. FIG. 2 is a schematic diagram of information regarding molecular structure. It is a schematic diagram of a character string. FIG. 2 is a schematic diagram of a circular structure and a corresponding string. FIG. 2 is a schematic diagram of a circular structure and a corresponding string. It is a schematic diagram explaining a conventional example. It is a flowchart explaining the 2nd processing example by an information processing device.

Embodiments will be described below with reference to the drawings.

As shown in FIG. 1, the information processing device 1 according to the present embodiment is one or more computers including, for example, a calculation section 10, a storage section 20, an input section 30, and an output section 40.

The calculation unit 10 performs various processes necessary for generating molecular descriptors of polymers, machine learning using molecular descriptors, predicting properties of polymers, and the like. Details of the processing contents will be described later. The calculation unit 10 includes, for example, one or more processors. The processor is, but is not limited to, a general-purpose processor or a dedicated processor specialized for specific processing. The various processes performed by the calculation unit 10 may be distributed to multiple processors for each process. In addition, in this embodiment, a "polymer" is a compound formed by polymerizing a plurality of monomers. Furthermore, in this embodiment, the "properties of a polymer" include, but are not limited to, the refractive index, elastic modulus, or glass transition temperature of the polymer.

The storage unit 20 stores information regarding the molecular structure of a polymer, a program for operating the information processing device 1, and the like. The storage unit 20 includes, for example, one or more memories. The memory is, for example, a semiconductor memory, a magnetic memory, or an optical memory, but is not limited to these.

The input unit 30 includes, for example, one or more input interfaces that can detect user input and send input information to the calculation unit 10. The input interface may be, for example, but not limited to, physical keys, capacitive keys, or a touch screen. The user can input information regarding the molecular structure of the polymer and the like via the input unit 30 .

The output unit 40 includes one or more output interfaces capable of outputting information including, for example, molecular descriptors and properties of the polymer to a user. The output interface is, for example, a display or the like, but is not limited thereto.

In the above configuration, the operation of the information processing device 1 is realized by executing a program stored in the storage unit 20 by a processor included in the calculation unit 10. When the information processing device 10 is composed of a plurality of computers connected to each other for data communication, the arithmetic processing may be distributed to the arithmetic units 10 of the plurality of computers, or the arithmetic processing may be distributed to the storage units 20 of the plurality of computers. Various information or programs may be stored in a distributed manner.

Next, with reference to FIG. 2, a first processing example by the information processing device 1 will be described.

In step S100, the calculation unit 10 reads information regarding the molecular structure of the polymer via the input unit 30. Information regarding the molecular structure of the polymer includes, for example, the compound name, molecular structural formula, CAS number, etc. of the polymer, and is stored in the storage unit 20 in a format such as a MOL file or SDF (Structure Data File). FIG. 3 shows the chemical structural formula of polyethylene terephthalate. In step S100, the information regarding the molecular structure of the polymer is stored in a format in which the molecular structure of the polymer is converted into a string using SMILES (Simplified Molecular Input Line Entry System) notation or SMARTS (SMILES arbitrary target specification) notation. It may be stored in the section 20.

In step S110, the calculation unit 10 converts the molecular structure of the polymer into a character string based on the information regarding the molecular structure of the polymer acquired in step S100. For example, the calculation unit 10 uses RDKit or the like stored in advance in the storage unit 20 to convert the molecular structure of the polymer into a character string written in SMILES notation or SMARTS notation. FIG. 4 is an example of a character string for polyethylene terephthalate written using the SMILES notation. However, the string conversion method in the present disclosure is not limited to this. Furthermore, in step S110, the molecular structure of the polymer may be graphed instead of converted into a character string. Note that in step S100, if the molecular structure of the polymer is converted into a character string using SMILES notation or SMARTS notation, the calculation unit 10 does not perform the process of step S110, but performs the processes of step S120 and subsequent steps.

In step S120, the calculation unit 10 determines the repeating unit of the polymer based on the molecular structure of the polymer converted into a character string in step S110. In the present embodiment, the "repeating unit" is a monomer constituting a polymer or a combination of a plurality of monomers, and is a pattern of strings that are repeated in the molecular structure of a string-formed polymer. For example, the repeating unit of the polymer is determined by the calculation unit 10 detecting a pattern of repeated character strings from the molecular structure of the polymer converted into a character string. The number of monomers included in the repeating unit is appropriately set in consideration of the molecular size of the polymer, the amount of calculation, the calculation time, and the like. The number of monomers is preferably 3 or more and 5 or less. When the number of monomers is three or more, the consistency of the atomic arrangement between the structure of the polymer composed of the monomers and the cyclic structure obtained by cyclizing the repeating units of the polymer is improved. If the number of monomers is 5 or less, the amount of calculation and calculation time required to generate a molecular descriptor, which will be described later, can be reduced compared to the case where the number of monomers is 6 or more. The number of monomers is more preferably 4 or more and 5 or less. Note that the number of monomers may be arbitrarily set in advance in the program executed by the calculation unit 10, or may be set appropriately by the user via the input unit 30.

In step S130, the calculation unit 10 generates a cyclic structure by cyclizing the repeating unit determined in step S120. Specifically, the calculation unit 10 generates a character string corresponding to a molecular structure in which terminal parts of repeating units are connected to each other. 5 and 6 show the respective cyclic structures and character strings when the number of monomers constituting polyethylene terephthalate is 1 to 5. In this embodiment, "to cyclize" means not to actually cyclize by a chemical reaction or the like, but to virtually connect the terminal parts included in the repeating unit on a computer program. 5 and 6 show a cyclic structure in which repeating units of polyethylene terephthalate are cyclized. FIG. 5 shows cyclic structures in which the number of monomers contained in the repeating unit is 1 to 3. FIG. 6 shows cyclic structures in which the number of monomers contained in the repeating unit is 4 to 5. Note that in FIGS. 5 and 6, atoms at the ends of each character string are virtually bonded to each other on a computer program. Hereinafter, step S130 will be explained in detail.

The degree of polymerization of polymers is generally as high as several hundred to several thousand. For this reason, if information regarding the molecular structure of a polymer is handled by a computer or the like, the amount of calculation and calculation time will increase accordingly. Therefore, for example, as shown in FIG. 7, it is possible to reduce the amount of calculation and calculation time by using a molecular descriptor corresponding to an oligomer formed by virtually polymerizing monomers constituting the polymer. However, when molecular descriptors corresponding to oligomers are used, terminal portions are inevitably generated. This tail end leads to a decrease in the accuracy of predicting the properties of the polymer. On the other hand, in this embodiment, since the repeating unit is cyclized, the influence of the terminal portion on the molecular descriptor is eliminated, and as a result, the prediction accuracy of the polymer properties is improved.

In step S140, the calculation unit 10 generates a molecular descriptor corresponding to the cyclic structure in step S130. For example, a molecular descriptor is generated by the calculation unit 10 using RDKit or the like stored in advance in the storage unit 20. The molecular descriptor generated in step S140 may be stored in the storage unit 20, or may be provided to a user or the like via the output unit 40. The method for generating molecular descriptors in step S130 will be described in detail below.

As shown in FIGS. 5 and 6, the cyclic structures in step S130 are all cyclic structures corresponding to polyethylene terephthalate, but have arbitrariness depending on the number of monomers contained in the repeating unit. From the viewpoint of further improving prediction accuracy, it is preferable to remove this arbitrariness. Therefore, in step S140, it is preferable to use MACCSkey, ECFP (Extended Connectivity Fingerprint), FCFP (Functional-class extended Connectivity Fingerprint), or the like. When using MACCSkey, the molecular descriptor includes information regarding whether or not 166 types of predefined partial structures are included in the cyclic structure in step S130. When using ECFP and FCFP, for each carbon atom included in the cyclic structure in step S130, the non-hydrogen atoms that are usually two or three adjacent from the carbon atom are treated as one partial structure. The molecular descriptor includes information regarding the type and number of this substructure.

To put Steps S100 to S140 in another way, the information processing device 1 according to the present embodiment determines a repeating unit based on information regarding the molecular structure of a polymer, and determines a molecule corresponding to a cyclic structure formed by cyclizing the repeating unit. Generate a descriptor. Therefore, according to this embodiment, a data structure including a molecular descriptor corresponding to a cyclic structure formed by cyclizing repeating units of a polymer is obtained. This data structure can be used as training data for the information processing device 1 to generate a learning model for predicting properties of the polymer by machine learning. In this embodiment, "teacher data" refers to a set of molecular descriptors and properties of a polymer, and "learning model" refers to a mathematical model that shows the correlation between the molecular structure and properties of a polymer.

Next, with reference to FIG. 8, a second processing example by the information processing device 1 will be described. In the second processing example, molecular descriptors corresponding to a plurality of known polymers generated in the same manner as in the first processing example are stored in advance in the storage unit 20. Further, information indicating the characteristics of a plurality of known polymers is stored in advance in the storage unit 20.

In step S200, the calculation unit 10 acquires information indicating a molecular descriptor and characteristics from the storage unit 20.

In step S210, the calculation unit 10 generates a learning model by performing machine learning on the correlation between the molecular structure and the properties of a plurality of known polymers, using pairs of molecular descriptors and properties as training data. . This machine learning uses any learning algorithm such as neural networks, support vector regression, random forests, LASSO (least absolute shrinkage and selection operator) regression, Ridge regression, Gaussian process regression, or PLS (Partial Least Squares) regression. be able to.

In step S220, the calculation unit 10 uses the learning model generated in step S210 to predict the properties of the polymer. Specifically, the calculation unit 10 generates, for example, a molecular descriptor corresponding to the molecular structure of a polymer whose properties are unknown, in the same manner as in the first processing example. Subsequently, the calculation unit 10 inputs the generated molecular descriptor into the learning model and predicts the properties of this polymer. Subsequently, the calculation unit 10 outputs the prediction result to the output unit 40.

To rephrase steps S200 to S220, the information processing device 1 according to the present embodiment performs machine learning on the correlation between the molecular structure and the properties using the set of the molecular descriptor and the polymer properties as training data. The information processing device 1 also predicts the properties of the polymer using a learning model generated by machine learning. According to the present embodiment, it is possible to improve the accuracy of predicting the properties of, for example, a polymer whose properties are unknown.

Although the present disclosure has been described above based on the drawings and embodiments, it should be noted that those skilled in the art can easily make various changes or modifications based on the present disclosure. It should therefore be noted that these variations or modifications are included within the scope of this disclosure. For example, functions included in each step can be rearranged so as not to be logically contradictory, and a plurality of steps can be combined into one or divided.

For example, a configuration in which a general-purpose computer functions as the information processing device 1 according to the embodiment described above is also possible. Specifically, a program that describes the processing content for realizing each function of the information processing device 1 according to the embodiment described above is stored in the memory of a general-purpose computer, and the program is read and executed by the processor of the general-purpose computer. let Therefore, the present disclosure can also be implemented as a program executable by a processor.

As examples, the invention example, comparative example 1, and comparative example 2 described below were prepared, and the prediction accuracy of polymer properties was verified.

As an invention example, using information on the molecular structure of a polymer provided by a polymer database (PoLyInfo), a cyclic structure (hereinafter, " A molecular descriptor was generated using MACCSKey.

As Comparative Example 1, a linear structure composed of one monomer (hereinafter referred to as "linear monomer") was generated using the same information on the molecular structure of the polymer as in the invention example, and using MACCSKey. generated molecular descriptors.

As Comparative Example 2, a linear structure composed of five monomers (hereinafter referred to as "linear pentamer") was generated using the same information on the molecular structure of the polymer as in the invention example, and using MACCSKey. generated molecular descriptors.

Same as the second processing example described above, using the molecular descriptors according to the invention example, comparative example 1, and comparative example 2 as explanatory variables, and using the refractive index of the polymer provided by the polymer database (PoLyInfo) as the objective variable. Machine learning was performed using support vector regression. Table 1 shows the coefficient of determination (500 times average value) under the conditions where the hyperparameters were optimized. Note that the coefficient of determination is a value for test data calculated using a 10-fold cross validation method, and means that the closer it is to 1, the better the prediction accuracy is.

As shown in Table 1, prediction accuracy was improved by predicting polymer properties using molecular descriptors corresponding to cyclic structures.

According to the present disclosure, a computer operating method and the like that can improve the prediction accuracy of polymer properties are provided.

1 Information processing device 10 Arithmetic unit 20 Storage unit 30 Input unit 40 Output unit

Claims (6)

The computer is
determining repeating units of the polymer based on information about the molecular structure of the polymer;
generating a molecular descriptor corresponding to a cyclic structure formed by cyclizing the repeating unit;
Machine learning the correlation between the molecular structure and the properties using the set of the molecular descriptor and the properties of the polymer as training data;
How it works, including: The operating method according to claim 1 , further comprising the step of predicting properties of the polymer based on information regarding the molecular structure of the polymer using the learning model generated by the machine learning. to the computer,
determining repeating units of the polymer based on information about the molecular structure of the polymer;
generating a molecular descriptor corresponding to a cyclic structure formed by cyclizing the repeating unit;
Machine learning the correlation between the molecular structure and the properties using the set of the molecular descriptor and the properties of the polymer as training data;
A program to run. 4. The program according to claim 3 , further causing the computer to execute a step of predicting properties of a polymer using the learning model generated by the machine learning. determining the repeating unit of the polymer based on information about the molecular structure of the polymer;
generating a molecular descriptor corresponding to a cyclic structure formed by cyclizing the repeating unit;
Machine learning the correlation between the molecular structure and the properties using the set of the molecular descriptor and the properties of the polymer as training data;
Information processing device. The information processing device according to claim 5 , wherein the learning model generated by the machine learning is used to predict the properties of the polymer based on information regarding the molecular structure of the polymer.

Images