JP7133534B2 - AUTOMATIC COMPOUND STRUCTURE GENERATOR, AUTOMATIC COMPOUND STRUCTURE GENERATION PROGRAM AND AUTOMATIC COMPOUND STRUCTURE GENERATION METHOD - Google Patents

Description

The present invention relates to an automatic compound structure generation apparatus, an automatic compound structure generation program, and an automatic compound structure generation method for automatically generating compound structures.

In drug discovery research, a new drug candidate compound (hereinafter referred to as a lead compound) that has been discovered as a new drug candidate is used as a standard, and by repeating design, synthesis, and evaluation so as to change the structure of the lead compound, the compound as a new drug is developed. An operation (hereinafter referred to as an optimization program) is performed to gradually optimize the structure.

In the optimization program, for example, a process of searching for new compounds derived from compounds described in SMILES notation using a method called ChemTS to which Monte Carlo Tree Search (MCTS) is applied is performed. there is At this time, while maintaining the main activity (drug effect) as a drug in the lead compound, it is possible to improve the compound's absorption, distribution, metabolism, excretion, and toxicity in humans and animals. Compounds are searched to improve the ADMET attribute that indicates the property.

https://arxiv.org/abs/1710.00616

By the way, when using ChemTS to search for new compounds from compounds described in SMILES notation, there is a problem that there are compound structures that cannot be expressed in SMILES notation, and the range of search by the optimization program is limited. .

One aspect of the present invention is an automatic compound structure generation apparatus for automatically generating a structure of a compound, wherein node information representing atoms constituting a basic structure of a compound as nodes, and bonding states of the atoms are Using supervised training data containing a combination of an adjacency matrix in a graph structure represented as edges connecting nodes and an additional structure known to be connected to any of the nodes of the basic structure, Automatic generation learning means for constructing an automatic structure generation model subjected to machine learning for automatically generating an additional structure that can bind to a basic structure, and inputting the basic structure of a target compound into the automatic structure generation model. , automatic structure generation means for automatically generating an additional structure that can be bound to the basic structure as an output of the automatic structure generation model;
A compound structure automatic generation device characterized by comprising

Here, the structure automatic generation model preferably uses a graph neural network (GNN).

Further, it is preferable that the structure automatic generation means automatically generates an additional structure that can be combined while maintaining the basic structure.

Another aspect of the present invention is an automatic compound structure generation program for automatically generating a structure of a compound, wherein a computer generates node information representing atoms constituting a basic structure of a compound as nodes, and bonds between the atoms. Using supervised training data including a combination of an adjacency matrix in a graph structure representing states as edges connecting the nodes, and an additional structure known to connect to any of the nodes of the basic structure , an automatic generation learning means for constructing an automatic structure generation model through machine learning for automatically generating an additional structure that can bind to the basic structure of the compound; and inputting the basic structure of the target compound into the automatic structure generation model. and automatic structure generation means for automatically generating an additional structure that can be bound to the basic structure as an output of the automatic structure generation model.

Another aspect of the present invention is an automatic compound structure generation method for automatically generating a structure of a compound, wherein a computer stores node information representing atoms constituting a basic structure of a compound as nodes, and bonds of the atoms. Using supervised training data including a combination of an adjacency matrix in a graph structure representing states as edges connecting the nodes, and an additional structure known to connect to any of the nodes of the basic structure , an automatic generation learning step of constructing an automatic structure generation model through machine learning for automatically generating an additional structure that can bind to the basic structure of the compound; and inputting the basic structure of the target compound into the automatic structure generation model. and an automatic structure generation step of automatically generating an additional structure that can be bound to the basic structure as an output of the automatic structure generation model.

One object of the embodiments of the present invention is to provide an automatic compound structure generation apparatus, an automatic compound structure generation program, and an automatic compound structure generation method that can automatically generate a new compound structure. Other objects of embodiments of the present invention will become apparent by reference to the specification as a whole.

BRIEF DESCRIPTION OF THE DRAWINGS It is a figure which shows the structure of the compound structure automatic generation apparatus in embodiment of this invention. 1 is a flow chart showing a compound structure automatic generation method according to an embodiment of the present invention. It is a figure which shows the example of the compound database in embodiment of this invention. It is a figure which shows the structural example of the compound in embodiment of this invention. It is a figure which shows the example of an adjacent row when the compound in embodiment of this invention is represented as graph structure. It is a figure which shows the example of an adjacent row when the compound in embodiment of this invention is represented as graph structure. It is a figure which shows the example of the basic structure and differential structure of the compound in embodiment of this invention. FIG. 4 is a diagram illustrating machine learning of the compound structure automatic generation model in the embodiment of the present invention;

The compound structure automatic generation device 100 according to the embodiment of the present invention comprises a processing unit 10, a storage unit 12, an input unit 14, an output unit 16 and a communication unit 18, as shown in FIG.

The compound structure automatic generation device 100 can be configured by a general computer. The processing unit 10 includes a CPU and the like, and integrally performs processing in the compound structure automatic generation apparatus 100 . The processing unit 10 executes the compound structure automatic generation program stored in the storage unit 12 to perform the compound structure automatic generation process according to the present embodiment. The storage unit 12 stores a compound structure automatic generation model (compound structure automatic generator) used in the compound structure automatic generation process, compound structure data that is the basis for drug discovery research, compound structure data obtained by automatic generation, etc. Stores information necessary for automatic compound structure generation processing. The storage unit 12 can be composed of, for example, a semiconductor memory, a hard disk, or the like. The storage unit 12 may be provided inside the compound structure automatic generation apparatus 100, or may be provided outside so as to be accessible from the processing unit 10 using an information network such as wireless or wired. The input unit 14 includes means for inputting information to the compound structure automatic generation device 100 . The output unit 16 includes means for displaying information processed by the compound structure automatic generation device 100 . The communication unit 18 includes an interface for exchanging information with an external device (server or the like). The communication unit 18 enables communication with external devices by being connected to an information communication network such as the Internet.

[Compound structure automatic generation processing]
The compound structure automatic generation process according to the present embodiment will be described below with reference to the flowchart of FIG. The compound structure automatic generation apparatus 100 executes a compound structure automatic generation program to perform machine learning for automatically generating a compound structure derived from a compound structure serving as a basis for automatic generation, thereby automatically generating a compound structure. A model (compound structure automatic generator) is generated, and a process of automatically generating a new compound structure using the compound structure automatic generation model is performed.

In the present embodiment, as an example, processing for automatically generating the structure of a derivative compound from the structure of a lead compound found as a new drug candidate in drug discovery research will be described. However, the scope of application of the present invention is not limited to this, and any process that derives a new compound structure from the structure of a basic compound is applicable.

In step S10, a process of reading training data for machine learning of the compound structure automatic generation model is performed. Through the processing in this step, the compound structure automatic generation device 100 functions as data acquisition means.

The structure of a compound can be represented as a graph structure. That is, it can be expressed as a graph structure in which "atoms" constituting a compound are "nodes" and "bonds between atoms" are "edges" connecting the nodes. Specifically, information indicating element species of atoms corresponding to "nodes" of the graph structure and bonding relationships and bonding states between atoms corresponding to "edges" is stored in the storage unit 12 as a compound database.

FIG. 3 shows examples of compound structures. FIG. 3(a) shows the structure of a compound, and FIG. 3(b) shows a graph structure in which node numbers are assigned to atoms constituting the compound.

FIG. 4 shows an example of structural data representing the structure of the compound stored as a compound database. In the compound database, a compound ID uniquely assigned to each compound, a compound name, element species for each node, adjacency matrix data indicating the bonding state between nodes, and combinable chemical structures are stored in association with each other.

Note that the element species for each node can be expressed as a feature vector for each node. For example, when the feature vector is represented by [carbon (C), hydrogen (H), oxygen (O), nitrogen (N)...], if the node is carbon (C), the feature vector [1, 0,0,0...], if the node is hydrogen (H) then the feature vector [0,1,0,0...] if the node is oxygen (O) then the feature vector [ 0,0,1,0...], or the feature vector [0,0,0,1...] if the node is nitrogen (N).

An adjacency matrix is a matrix that indicates how nodes are connected by edges in a graph structure. An adjacency matrix is a square matrix with rows and columns of the number of nodes in the graph structure.

For example, in the configuration of the compound shown in FIG. 3A, the mutual coupling relationships of nodes 1 to 7 can be expressed as an adjacency matrix shown in FIG. That is, in the adjacency matrix, when the atom corresponding to the node N and the atom corresponding to the node M are chemically bonded, the values of the matrix element (N, M) and the matrix element (M, N) are "1". and the value is "0" when not chemically bonded. The value of the diagonal element is "0". For example, since the atom corresponding to node 1 and the atom corresponding to node 2 are chemically bonded, the matrix element (1,2) and the matrix element (2,1) are set to "1", and the atom corresponding to node 2 and Since atoms corresponding to node 3 are not chemically bonded, matrix elements (2,3) and (3,2) are "0".

When constructing an automatically generated model of a compound including the state of bonding between nodes, an adjacency matrix representing the state of chemical bonding as well, as shown in FIG. 6, may be used. For example, when the atom corresponding to node N and the atom corresponding to node M are chemically bonded, the adjacency matrix includes matrix element (N, M) and matrix element (M, N ) can be expressed as "1", double bond as "2", triple bond as "3", and not chemically bonded as "0". The value of the diagonal element is "0".

An existing graph analysis program can be applied to express the structure of a compound as a graph structure.

In the present embodiment, in order to perform machine learning on a compound structure automatic generation model, graph structure data indicating the composition of a compound and chemical structures that can be derived in combination in the compound are combined as additional structures and used as supervised training data. . For example, as shown in FIGS. 7(b) and 7(c), for a compound having the basic structure of FIG. 7(a), any atom (node in the graph structure) of the basic structure can If such an additional structure can exist, the basic structure and the chemical structures (CM1, CM2), which are the additional structures, are combined and used as supervised training data. Specifically, a basic structure may be used as training data, and an additional structure that can be combined with the basic structure may be associated as teacher data and stored in advance in a compound database as supervised training data.

In step S12, machine learning of the structure automatic generation model of the compound is performed. Through the processing in this step, the compound structure automatic generation device 100 functions as automatic generation learning means.

As shown in FIG. 8, the processing unit 10 converts, as supervised training data, a combination of the basic structure of the compound of the supervised training data acquired in step S10 and the additional structure that can be derived from the compound, By inputting the structure of a new compound, a compound structure automatic generation model is machine-learned so that an additional structure that can be derived and combined with the compound is output.

A graph neural network (GNN) is preferably applied to the compound structure automatic generation model. GNN is a neural network that handles graph structures, and many models have been proposed. In the machine learning model, it is preferable to appropriately select the number of layers of the neural network, the activation function, the loss function, and the like.

Specifically, for example, the processing is performed as follows. Using the basic structure of the compound obtained in step S10 as an input to the GNN, graph convolution processing (Convolution) is performed on the feature vector and the adjacency matrix indicating the element species for each node of the basic structure to obtain the feature vector for each node. A feature vector for each node in the graph structure of the basic structure thus obtained is used as an input, and an automatic compound structure generation model is constructed that outputs a graph structure of additional structures that can be derived and combined with the basic structure. machine learning to

In machine learning, any basic structure included in the compound database and an additional structure that can be combined with the basic structure may be used as verification data (validation data) or evaluation data (test data).

In step S14, automatic generation processing of the compound structure is performed. Through the processing in this step, the compound structure automatic generation device 100 functions as structure automatic generation means.

The user inputs the basic structure of a compound for which a new derivative structure is to be automatically generated into the compound structure automatic generation model built in step S12, thereby creating an additional structure that can be combined with the basic structure. Generate automatically. Specifically, by obtaining the graph structure of the basic structure of the compound using a graph analysis program and inputting the graph structure into the compound structure automatic generation model, the additional structure that can be combined with the compound is automatically generated. generated and output.

As described above, in the compound structure automatic generation apparatus 100 according to the present embodiment, by inputting a basic structure that is the skeleton of a compound, while maintaining the basic structure that is the skeleton, the atoms ( It is possible to automatically generate an additional structure that can be connected to a node in the graph structure. Therefore, by combining the automatically generated addition structure and the basic structure, it is possible to automatically generate a new compound derived from the basic structure.

For example, in drug discovery research, by automatically generating an additional structure that can be combined with the basic structure of a lead compound, compounds derived from the lead compound can be obtained as candidates for new research. Specifically, for example, in an optimization program for drug discovery research, the ADMET attribute is improved by applying a compound derived from the lead compound obtained by the compound structure automatic generation device 100 to the MCTS method such as ChemTS. compounds can be searched for.

However, compounds to be automatically generated in the compound structure automatic generation device 100 are not limited to compounds in drug discovery research. Also, the method of using the compound obtained by the automatic compound structure generation apparatus 100 is not particularly limited.

In addition, in the compound structure automatic generation device 100 in the present embodiment, the data acquisition means, the automatic generation learning means, and the automatic structure generation means are realized by one device. You may make it implement|achieve in a main body. For example, some of these means may be implemented by a computer.

10 processing unit, 12 storage unit, 14 input unit, 16 output unit, 18 communication unit, 20 processing unit, 22 storage unit, 24 input unit, 26 output unit, 28 communication unit, 100 compound structure automatic generation device.

Claims (5)

A compound structure automatic generation device for automatically generating a compound structure,
Node information representing atoms constituting a basic structure of a compound as nodes, an adjacency matrix in a graph structure representing the bonding state of the atoms as edges connecting the nodes, and a connection to any of the nodes of the basic structure Using supervised training data containing combinations of additional structures known to automatically generated learning means;
automatic structure generation means for automatically generating an additional structure capable of binding to the basic structure as an output of the automatic structure generation model by inputting the basic structure of a target compound into the automatic structure generation model;
A compound structure automatic generation device comprising: The compound structure automatic generation device according to claim 1,
A compound structure automatic generation device, wherein the structure automatic generation model uses a graph neural network (GNN). The compound structure automatic generation device according to claim 1 or 2,
An apparatus for automatically generating a compound structure, wherein the automatic structure generating means automatically generates a bondable additional structure while maintaining the basic structure. A compound structure automatic generation program for automatically generating a compound structure,
the computer,
Node information representing atoms constituting a basic structure of a compound as nodes, an adjacency matrix in a graph structure representing the bonding state of the atoms as edges connecting the nodes, and a connection to any of the nodes of the basic structure Using supervised training data containing combinations of additional structures known to automatically generated learning means;
automatic structure generation means for automatically generating an additional structure capable of binding to the basic structure as an output of the automatic structure generation model by inputting the basic structure of the target compound into the automatic structure generation model;
A compound structure automatic generation program characterized by functioning as A compound structure automatic generation method for automatically generating a compound structure,
to the computer,
Node information representing atoms constituting a basic structure of a compound as nodes, an adjacency matrix in a graph structure representing the bonding state of the atoms as edges connecting the nodes, and a connection to any of the nodes of the basic structure Using supervised training data containing combinations of additional structures known to an automatically generated learning process;
an automatic structure generation step of automatically generating an additional structure capable of binding to the basic structure as an output of the automatic structure generation model by inputting the basic structure of the target compound into the automatic structure generation model;
A compound structure automatic generation method characterized by executing

Images