Polymer graph neural network and its realization method

Abstract

The present invention provides a computer-implemented method for predicting characteristic information of a polymer from its chemical structure. The present invention provides a method for graphically representing the chemical structure of a polymer, and a method and system for calculating characteristic information of the polymer from graph information of the polymer by machine learning the chemical structure of the polymer based on information defining the mutual connection relationships between each atom constituting the repeating unit of the polymer and the connection nodes to which the repeated repeating units are connected.

Description

[Technical field]

[0001] The present invention relates to a method and system for representing a polymer structure in a manner that can be recognized by a computer device. [Background technology]

[0002] A wide variety of methods have been proposed to represent the structure of a compound in a way that the computer can recognize it in order to input the structure of the compound into a computer for processing. Typical methods include SMILES, which represents the structure of a compound as a character string, or a method in which a character string converted into SMILES is represented by a descriptor.

[0003] However, for polymers that are not monomers, it is difficult to represent them using the SMILES method, so BigSMILES and hierarchical descriptors have been proposed.

[0004] However, such conventional techniques have problems in that the larger the molecule, the more complex the representation becomes, and the techniques are limited to representing polymers in a fragmentary manner. There is no technique capable of processing molecules based on such representation, and the characteristics of polymers, in which repeating units are repeated infinitely, cannot be represented properly.

[0005] The related prior art is as follows: [Prior art documents] [Patent documents]

[0006] [Patent Document 1] Republic of Korea Patent Publication No. 2021-0042777 [Patent Document 2] Republic of Korea Patent Publication No. 2021-0110539 Summary of the Invention [Problem to be solved by the invention]

[0007] In order to solve the above-mentioned problems, the present invention aims to provide a data structure that represents a polymeric substance using graphic information, and further to provide a method and system for predicting a predetermined property of the polymer using the data structure that represents the graphic information of the polymer. [Means for solving the problem]

[0008] In order to solve the above-mentioned problems of the prior art, the present invention provides a data processing method for a polymer chemical structure, comprising: a polymer node connection variable setting step of setting, as a polymer node connection variable, a connection relationship between nodes representing each atom constituting a repeating unit of a selected polymer and a connection node connected to a connection point, which is a point where the repeating unit is repeatedly connected, among the nodes; a polymer graph node connection information assignment step of assigning an attribute value of the connection relationship between the nodes and the connection nodes to the set polymer node connection variable; and a polymer graph node connection information assignment step of assigning an attribute value of the connection relationship between the nodes and the connection nodes to the set polymer node connection variable, in a method for converting a chemical structure of a polymer into computer-processable data.

[0009] The data processing method for polymer chemical structures of the present invention may further include a polymer node attribute variable setting step of setting attribute information of nodes corresponding to constituent atoms of a repeating unit forming a selected polymer and information of connection nodes connected to connection points at which the repeating units are repeatedly connected among the nodes as polymer node attribute variables, and a polymer graph node attribute information assignment step of assigning attribute values ​​of the nodes and connection nodes to the set polymer node attribute variables. The data processing method for polymer chemical structures of the present invention may further include a polymer node attribute variable setting step of setting attribute information of nodes corresponding to each atom forming a repeating unit forming a selected polymer and information of connection nodes connected to connection points at which the repeating units are repeatedly connected among the nodes as polymer node attribute variables, a polymer graph node attribute information assignment step of assigning attribute values ​​of the nodes and connection nodes to the set polymer node attribute variables, a polymer edge attribute variable setting step of setting information of edges connecting nodes corresponding to each atom forming a repeating unit forming a selected polymer and connection edges connecting at least one of the nodes to the connection node as polymer edge attribute variables, and a polymer graph edge attribute information assignment step of assigning attribute values ​​of the edges and connection edges to the set polymer edge attribute variables.

[0010] The present invention also provides a method for generating a computer-implemented model for graphically describing the chemical structure of a selected polymer and analyzing a predetermined characteristic of the polymer, the method comprising: a basic data acquisition step comprising chemical structure information of a number of training polymers and known values ​​of predetermined characteristic information of the polymer; a data pre-processing step for converting the chemical structure information of the training polymer, among the basic data, into polymer graph information; a polymer characteristic information prediction artificial neural network configuration step for configuring a polymer characteristic information prediction artificial neural network to receive input of the polymer graph information and output a predicted value of the predetermined characteristic information of the polymer; a training data input step for inputting the polymer graph information of the training polymer and the predetermined characteristic information of the training polymer into the polymer characteristic information prediction artificial neural network; and a parameter setting step for setting the parameters of the polymer characteristic information prediction artificial neural network based on a comparison result between the predicted value of the predetermined characteristic information of the training polymer output by the polymer characteristic information prediction artificial neural network and the known value of the predetermined characteristic information of the training polymer. and a learning step of a polymer characteristic information prediction artificial neural network that updates a learning parameter of the polymer characteristic information prediction artificial neural network to update a learning parameter of the polymer characteristic information prediction artificial neural network, wherein the polymer graph information includes at least one of the following: polymer graph node connection information, which is information representing a connection relationship between nodes corresponding to each atom constituting the repeating unit forming the polymer and connection nodes connected to connection points, which are points at which the repeating units are repeatedly connected among the nodes; polymer node attribute information, which is information representing an attribute of a node corresponding to each atom constituting the repeating unit forming the polymer and an attribute of a connection node connected to a connection point, which is a point at which the repeating units are repeatedly connected among the nodes; and polymer edge attribute information, which is information representing an attribute of an edge connecting the nodes corresponding to each atom constituting the repeating unit forming the polymer and a connection edge connecting at least one of the nodes to the connection node.

[0011] The present invention further provides a computer-implemented method for graphically describing a chemical structure of a selected polymer and analyzing a predetermined property of the polymer, comprising: a polymer property information calculation artificial neural network construction step for generating an artificial neural network machine-learned to predict a predetermined property of the selected polymer based at least in part on chemical structure data associated with the selected polymer; a polymer graph information acquisition step for acquiring polymer graph information describing the chemical structure of the selected polymer by graphical data; a polymer graph input step for providing the acquired polymer graph information as an input of the machine-learned graph neural network; and a step of receiving prediction data describing the predetermined property of the selected polymer as an output of the machine-learned graph neural network, the step of: The polymer graph information describing the chemical structure of a molecule using graph information includes at least one of the following: polymer graph node connection information, which is information representing a mutual connection relationship between nodes corresponding to each atom constituting the repeating units forming the polymer and connection nodes connected to connection points, which are points at which the repeating units are repeatedly connected among the nodes; polymer node attribute information, which is information representing attribute information of nodes corresponding to each atom constituting the repeating units forming the polymer and attributes of connection nodes connected to connection points, which are points at which the repeating units are repeatedly connected among the nodes; and polymer edge attribute information, which is information representing attributes of edges connecting the nodes corresponding to each atom constituting the repeating units forming the polymer and connection edges connecting at least one of the nodes to the connection node.

[0012] In this case, the step of providing the acquired polymer graph information as an input of the machine-learned graph neural network includes a polymer adjacency matrix constructing step of constructing an adjacency matrix representing a mutual connection relationship between two or more nodes representing each atom constituting the single molecule, and a polymer adjacency matrix input step of inputting the constructed graph matrix as the graph neural network, and the polymer adjacency matrix may further include connection relationship information between the connecting nodes and the nodes. Effect of the Invention

[0013] According to the present invention, by providing a data structure capable of representing a polymer by graphic information usable by a computer, it is possible to realize a method and system for predicting characteristic information of a polymer substance by machine learning the relationship between the molecular structure and characteristic information of the polymer substance using a graph neural network (GNN) and a message passing neural network (MPNN). In addition, it is possible to provide a method and system for predicting polymer characteristic information with improved accuracy compared to conventional techniques.

[0014] The drawings attached to this specification are intended to illustrate preferred embodiments of the present invention and serve to facilitate a further understanding of the technical ideas of the present invention as well as the content of the invention, and therefore the present invention should not be interpreted as being limited to only the matters depicted in the drawings. [Brief description of the drawings]

[0015] [Figure 1] 1(a) and 1(b) are diagrams showing graphic representations of monomers and polymers according to the prior art. [Diagram 2] 1(a) and (b) show a graphical representation of a polymer according to the present invention. [Diagram 3]1A and 1B are exemplary diagrams illustrating a graphical representation of a polymer according to the present invention. [Figure 4] 1(a) and 1(b) are diagrams showing the progression of learning based on message passing through a graphical representation of a polymer according to the present invention. [Diagram 5] 1(a) and 1(b) are diagrams illustrating a procedure for realizing an artificial neural network for calculating characteristic information of a polymer using a graphical representation of the polymer according to the present invention. [Figure 6] FIG. 1 is a diagram showing a polymer characteristic information prediction system according to the present invention. [Figure 7a] FIG. 1 is a diagram showing an example of polymer graph information generation according to the prior art and the present invention. [Figure 7b] FIG. 1 is a diagram showing an example of polymer graph information generation according to the prior art and the present invention. [Figure 8a] FIG. 2 is a diagram showing another example of polymer graph information conversion according to the prior art and the present invention. [Figure 8b] FIG. 2 is a diagram showing another example of polymer graph information conversion according to the prior art and the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0016] 1. Graphical representation of polymers according to the present invention

[0017] First, the method of graphically representing a polymer according to the present invention will be described with reference to FIGS.

[0018] 1(a) and (b) are diagrams showing the structures of different polymers and the monomers that form each polymer. However, when attempting to process molecular structure data using an artificial neural network, particularly when attempting to represent the molecular structure as a graph for application to a GNN, there has been no effective method for representing the chemical structure of a polymer as a graph.

[0019] Although the monomers forming the polymers shown in Figures 1(a) and (b) are different from each other, they are the same except for some of the ends of the monomers, and therefore their monomer graphs are expressed as being very similar. Therefore, it is inaccurate to express the graph of the polymer simply as a graph of the monomers, and to express the polymer graph as a repetition of the monomer graphs.

[0020] For this reason, the present invention proposes a novel polymer graph representation method as shown in Fig. 2. The polymer graph information representation method according to the present invention will be described below with reference to Fig. 2.

[0021] First, in the graph representation of the polymer of the present invention, monomers constituting the polymer and repeating units which are repeated to form the polymer are derived. FIG. 1 shows an example of a conventional polymer representation and a display of monomers constituting the polymer, and FIG. 2 shows a repeating unit constituting the polymer of FIG. 1. In the repeating unit of FIG. 2, for the representation of the polymer of the present invention, each atom is represented as a node 10, and the bond between the atoms is represented as an edge 20. Next, when the repeating units are repeated to form a polymer, which is one of the features of the present invention, the point where the repeating units are repeatedly connected is represented as an attachment point 30, and the node to which the attachment point is connected is represented as a "connection node". Also, the attachment point 30 is a virtual representation of a connection node of adjacent repeating units which are repeatedly connected, and the connection connecting the attachment point and the connection node is represented as an attachment edge 40.

[0022] According to the polymer graph representation method of the present invention, when learning an artificial neural network model based on the chemical structure of a polymer, particularly when learning is performed by applying an artificial neural network model such as MPNN (message passing neural network) or GNN (graph neural network) based on message passing, learning in which information of the opposite node to which the monomer is connected is passed can be performed by adding only a minimum amount of information in addition to the monomer structure information, and learning that reflects the unit repetition characteristics of the polymer can be performed.

[0023] FIG. 4 shows an example of structural information of monomers and polymers represented in the form of a binary tree. In addition to the monomer binary tree in FIG. 4(a), a connection point tree is added as shown in FIG. 4(b), which shows that information on the opposite node is included.

[0024] 2. Converting the graph representation of the polymer according to the present invention into computer-usable data

[0025] In the present invention, the chemical structure of a polymer is converted into a data structure in a form usable for computer processing, i.e., "graph data" by using a graph representation method that graphically represents the molecular structure or chemical structure of the polymer. In the present invention, this is called "polymer graph data".

[0026] The polymer graph information according to the present invention, which describes the chemical structure of a given or selected polymer using graphic information, includes at least one of the following information: nodes 10 representing two or more nodes each representing an atom constituting a repeating unit of a monomer forming the polymer, node data which is data associated with the nodes 10, edge data which is data associated with one or more edges 20 representing bonds between the respective nodes, attaching node data which is data associated with at least one connecting node among the nodes, which is a point where the repeating units forming the polymer are repeatedly connected, and attach edge data which is data associated with an attach edge 40 which is a connection connecting the attach point 30 and the connecting node.

[0027] The difference between the monomer graph information and the polymer graph information of the present invention is that the polymer graph information of the present invention represents the node 10 connected to the connection point 30 and its edge information by a connection node and a connection edge in expressing the graph information of a repeating unit including the monomer graph information.

[0028] 2.1. Adjacency matrix and feature matrix

[0029] When the graph structure of a monomer is converted into computer-usable data, a wide variety of data representation methods are possible, but typically, the edge information and node information can be represented by an adjacency matrix and a property matrix, respectively. The edge information is information related to bonds between atoms, and the node information is information related to each atom.

[0030] According to the polymer representation method of the present invention, in addition to the conventional monomer graph representation method, information on the nodes and connection edges 40 connected to the connection points 30 can be further added to convert the polymer graph into data that can be recognized by a computer.

[0031] (1) Representation of the adjacency matrix of a monomer graph

[0032] For the sake of simplicity, taking the simple structure shown in FIG. 3 as an example, the adjacency matrix A of a monomer repeat unit consisting of nodes 1, 2, 3, and 4 can be expressed as follows:

[0033]

number

[0034] The adjacency matrix A has rows and columns corresponding to the number of nodes, and each component A- i,j is information that indicates whether the i (i = 1, ..., 4)th node is connected to the j (j = 1, ..., 4)th node. The value of each matrix element indicates the connection information with other nodes. For example, node 1 is connected only to node 2, and is not connected to nodes 1, 3, and 4, so A of the adjacency matrix 1,j It can be expressed as (j=1, …,4)=[0 1 0 0].

[0035] (2) Representation of the adjacency matrix of a polymer graph according to the present invention

[0036] According to the polymer graph information representation method of the present invention, the polymer adjacency matrix PA, which is an expression of the adjacency matrix of the polymer graph illustrated in FIG. 3, can be expressed as follows:

[0037]

number

[0038] As shown in FIG. 3, two connection points 5 and 6 are added to the polymer adjacency matrix PA, which respectively represent nodes 4 and 1 of the adjacent monomer, so that nodes 1 and 4 are expressed as being connected to nodes 4 and 1, respectively. That is, in the monomer, there is no node to which nodes 1 and 4 are connected, so the matrix value showing the connection relationship between node 1 and node 4 is A. 1,4 =0, A 4,1 = 0, but in the polymer representation method of the present invention, nodes 1 and 4 are connected to nodes 4 and 1 of the adjacent monomers, respectively, so that PA 1,4 =1, P.A. 4,1 = 1, and contains the repeated connection information of the polymer. That is, nodes 1 and 4, which were not connected in the monomer, are connected to each other in the polymer, so they are represented as being connected to each other in the polymer adjacency matrix that displays the connection edge information.

[0039] However, in this case, since nodes 1 and 4 are not connected to each other within the monomer, in order to indicate that it is not a connection within the monomer but a connection with a node of an adjacent repeating unit, the node connection relationship can be expressed as a negative number and the polymer adjacency matrix can be expressed as shown below, or a "connecting node" characteristic that is connected to the connection point can be given to the characteristic of the node in the node characteristic matrix described below, or an adjacent repeating unit connection attribute can be given to the edge characteristic matrix.

[0040]

number

[0041] To explain again, the graph information, which graphs the chemical structure of the polymer according to the present invention and describes it with graphical data, includes, in addition to the graph information of the monomer repeating units, connection node information representing connection nodes, which are nodes to which the monomers are repeatedly connected, and information on connection edges 40 connected to the connection nodes.

[0042] According to the expression method of the polymer adjacency matrix PA, a monomer repeat unit composed of nodes 1, 2, 3, and 4 is represented as being connected to nodes 4 and 1 of adjacent repeat units at nodes 1 and 4, respectively, and repeated. However, in this case, since nodes 1 and 4 in Fig. 3 are not connected to each other within the monomer but to nodes of adjacent repeat units, these nodes can be designated as "connection nodes" and distinguished by giving a connection node attribute to the node characteristic matrix or an adjacent repeat unit connection attribute to the edge characteristic matrix.

[0043] This can be used as an embedding procedure to convert the chemical structure of a polymer into computer-processable data, and is achieved by using a polymer edge variable setting step of setting the mutual connection relationship between the nodes representing each atom constituting the monomer repeat unit of the selected polymer and the connection node at the point where the monomer is repeatedly connected as a polymer connection edge variable, and a polymer graph edge information assignment step of assigning the attribute value of the mutual connection relationship between the nodes and the connection node to the set polymer edge variable. As is clear from the above polymer adjacency matrix PA, the attribute value of the mutual connection relationship between the nodes and the connection node can be determined as "1" or "0" indicating connection / non-connection, and the polymer adjacency matrix is ​​a matrix representation of this. The polymer adjacency matrix of the present invention has its component values ​​like a normal adjacency matrix, but is different from the above by adding a connection node to a normal single molecule graph node as described above.

[0044] (3) Characteristic matrix of the monomer graph

[0045] The graph representation of a compound may have, as graph information, a property matrix representing predetermined attribute information of each node and predetermined attribute information of each edge, in addition to the adjacency matrix indicating the connection relationship between nodes. The monomer node property matrix having attribute value information (exemplarily, three attribute information) of each atom in the exemplary monomer graph of FIG. 3 may be expressed as follows:

[0046]

number

[0047] The node characteristic matrix NF has rows corresponding to the number of nodes and columns corresponding to the number of types of characteristic values ​​to be expressed for each node, and can represent a predetermined attribute to be expressed for each atom. i,j has the j-th property value of the i-th node. For the sake of explanation, the monomer property matrix NF is described as an example in which each node has three arbitrary attribute values.

[0048] The monomer graph may be represented by an edge property matrix EF, where each row represents an edge of the monomer and the column data of each row represents a predetermined attribute value of each edge. The edge property values ​​may include the type of chemical bond (single bond, double bond, etc.), the presence or absence of a ring, the presence or absence of conjugation, etc. For example, in the case of the monomer of FIG. 3, there are three connecting edges, so the edge property matrix has three rows, and each column is assigned an edge attribute value to accommodate the information. An example of an edge property matrix with three arbitrary edge attribute values ​​is as follows:

[0049]

number

[0050] (4) Characteristic matrix of the polymer graph according to the present invention

[0051] The polymer graph information according to the present invention may have additional attribute values ​​in addition to the attribute values ​​of the monomer node feature matrix. For example, the polymer node feature matrix PF (Polymer Feature matrix), which is the feature matrix of the exemplary polymer graph of FIG. 3, may also have additional attribute information in the third column as shown in the following example, which may be expressed as an attribute value indicating whether each node is a connection node or not. For example, the following polymer node feature matrix PNF may have connection node attribute information in the fourth column as shown below, in addition to the monomer node feature matrix NF containing the three types of attribute information of the four nodes. In the example of FIG. 3, nodes 1 and 4 are connection nodes connected to adjacent repeat units, and therefore have a value of "1", and other nodes are represented as having a value of "0" in the fourth column.

[0052]

number

[0053] This is accomplished by using a polymer node variable setting step of setting each atom of the monomer forming the selected polymer and a connection node where the monomer is repeatedly connected as a polymer node variable, and a polymer graph node information assignment step of assigning connection nodes connected to the set polymer node variables and assigning node attribute values ​​to each node. The node attribute values ​​assigned to the nodes may include atomic number, hybridization (SP3, SP2, etc.), number of hydrogens, number of electrons, presence or absence of rings, etc., and in one embodiment of the present invention, a connection node attribute value may be an attribute value of the presence or absence of a node connected to an adjacent repeat unit value.

[0054] In another embodiment of the present invention, the connection node information may be represented by an edge property matrix. For example, instead of or in addition to the node attribute value indicating whether or not a node is a connection node as in the previous polymer node property matrix, an attribute value indicating the presence or absence of a connection edge may be represented in the edge property matrix. In the following example showing FIG. 3(b), it can be seen that data in the fourth and fifth rows are added to accommodate information on the edges connecting connection nodes 1 and 4, i.e., edges 5_1 and 4_6 in FIG. 3(b), and an attribute information column indicating the presence or absence of a connection edge is added to the fourth column.

[0055]

number

[0056] That is, according to the present invention, the method of graphically representing the structure of a polymer and converting it into data that can be used by a computer can be selected from any one or more of the following methods.

[0057] 1) In an adjacency matrix representing the connection relationship of each node of a repeating unit consisting of at least one monomer, the components of a "connection node" connected to an adjacent repeating unit are assigned as a "connection" attribute.

[0058] 2) A "connection presence / absence" attribute field is provided to a node attribute matrix representing the attributes of each node of the repeating unit, and "connection" is assigned to the "connection presence / absence" attribute field of the "connecting node."

[0059] 3) The edge connected to the connection node is added to the connection edge in the edge attribute matrix of the repeating unit, a "connection presence / absence" attribute field is given to the edge attribute, and "connection" is assigned to the corresponding field of the connection edge.

[0060] 2.2. Method for converting the chemical structure of polymers into computer-processable data

[0061] In the present invention, in order to train an artificial neural network model as described below and to predict the properties of a polymeric substance using the trained artificial neural network, the chemical structure of the polymeric substance must be converted into data that can be processed by a computer. The data is converted into data according to the method of the present invention as described below.

[0062] (1) Repeating unit information acquisition step

[0063] This is a step of acquiring repeating unit information, which is a unit in which a monomer is repeated to form a polymer. In order to convert connection node information in the repeating unit, which is one of the features of the present invention, repeating unit information that forms a polymer is acquired. The repeating unit information is a step of determining whether a repeating unit that is repeated to form a polymer is repeated as a single monomer or as a bond of a single monomer, and describing the repeating unit by a graphical representation. The repeating unit may be a single monomer as shown in FIG. 2(a) or two monomers connected as shown in FIG. 2(b). In the case shown in FIG. 2(b), an already adjacent node becomes an attachment point.

[0064] (2) Polymer graph node connectivity information assignment step

[0065] In the present invention, the method for processing polymer graph information includes a step of setting, as variables, connection relationships between nodes representing each atom constituting a repeating unit of a polymer selected as an analysis target and connection nodes connected to connection points, which are points where the repeating units are repeatedly connected, among the nodes.

[0066] The attribute values ​​of the mutual connection relationships between the nodes and the connection nodes are assigned to the polymer node connection variables thus set. The polymer adjacency matrix given as an example above can be used to assign the attribute values ​​of each connection relationship to the polymer node connection variables.

[0067] (3) Polymer graph node attribute information assignment step

[0068] In addition, the polymer graph information processing method of the present invention may include a polymer node attribute variable setting step of setting attribute information of nodes corresponding to each atom constituting a repeating unit forming a selected polymer and attribute information of connection nodes connected to connection points, which are points where the repeating units are repeatedly connected, among the nodes, as polymer node attribute variables.

[0069] The method further includes a polymer graph node attribute information assignment step of assigning attribute values ​​of the nodes and connecting nodes to the polymer node attribute variables thus set. The node attribute information has an attribute value indicating whether the node is a connecting node or not, and a connecting node attribute value indicating that the node is connected to a connecting node of an adjacent repeat unit in the polymer is given as the node attribute information of the connecting node, unlike a node that is not a connecting node.

[0070] (4) Assignment of polymer graph edge attribute information

[0071] Furthermore, the polymer graph information processing method of the present invention may include a polymer edge attribute variable setting step of setting information of edges connecting nodes corresponding to each atom constituting a repeating unit forming a selected polymer and a connection edge connecting at least one of the nodes to the connection node as polymer edge attribute variables. A connection edge is an edge that connects a connection node to another node.

[0072] Furthermore, the present invention has a polymer graph edge attribute information assignment step for assigning attribute values ​​of the edges and connection edges to the set polymer edge attribute variables, and the connection edges are assigned a connection edge attribute value, which means that the connection node is connected to another node, as its edge attribute variable.

[0073] <Examples and Comparative Examples>

[0074] An example of expression of the above-mentioned polymer graph information according to the present invention will be described.

[0075] FIG. 7a shows a method of expressing the molecular structure of polyethylene terephthalate (PET), an example of a polymer, by graph information according to the conventional technology, and FIG. 7b shows a method of expressing polymer graph information according to the present invention.

[0076] In the conventional method, only the unit structure of a polymer is converted into a graph as shown in Fig. 7a. In this case, the atomic number, the presence or absence of a ring, and the number of hydrogen atoms are used as the node attribute values ​​of the node characteristic matrix. The edge characteristic matrix has the type of bond, whether it is aromatic, and whether it is conjugated as the edge attribute values.

[0077] In contrast, the graph information representation method according to the present invention further includes connecting node attribute information as a node attribute value of the node characteristic matrix, as shown in Fig. 7b, and in the example, since node 1 and node 14 are connected, the connecting node attribute information value is "1". The edge characteristic matrix further includes edge attribute values ​​as connecting edge attribute information, and the connecting edge "14-1" has attribute information "1" as a connecting edge.

[0078] FIG. 8 shows an example of converting the molecular structure of polyvinyl benzyl chloride (PVBC) as another polymer into graph information.

[0079] FIG. 8a shows the adjacency matrix, node property matrix, and edge property matrix represented by the prior art method.

[0080] FIG. 8b shows an example of the method according to the present invention, in which the molecular structure of PVBC is converted into graph information. Compared with the method according to the prior art of FIG. 8a, the node characteristic matrix further includes connection node attribute information. In the example, nodes 1, 2, 11, and 12 are connection nodes, and the node attribute information values ​​of these nodes are set to "1". When an already adjacent node of a repeat unit is an attachment point, as in PVBC, a repeat unit in which two monomers are connected is used to convert into graph information. This is the case described in relation to FIG. 2(b). As shown in the figure, the edge characteristic matrix further includes connection edge attribute information, and the connection edge attribute information of the connection edges "2-11" and "1-12" is set to "1".

[0081] 3. Method for analyzing polymer properties using the graph representation of the polymer according to the present invention

[0082] In the following, a method for processing polymer data in an artificial neural network according to the above-mentioned polymer graph representation method of the present invention will be described.

[0083] The present invention provides a method for constructing an artificial neural network model that analyzes specific properties of a polymer by applying the above-mentioned graph representation method of the polymer, and calculating the specific properties of the polymer through machine learning.

[0084] 3.1. Method for generating polymer property analysis model

[0085] 5(a) shows a method for generating a polymer property analysis model according to the present invention. Based on this, a method for generating a computer-implemented polymer property analysis model for predicting specific property information of a selected polymer from the chemical structure of the polymer according to the graph information representation method of the present invention will be described.

[0086] (1) Training data preparation process (T10)

[0087] First, the method includes a learning data preparation process (T10) for converting the chemical structure information of a large number of learning polymers, whose given characteristic information is known, into the polymer graph information of the present invention by the above-mentioned method from basic data composed of the chemical structure information of the learning polymers and the given known characteristic information of the polymers. This corresponds to a data pre-processing process of the basic data. The learning data thus prepared may be composed of a large number of data sets, each set consisting of {learning polymer graph information, known given characteristic information of the polymer}. Here, the known given characteristic information of the polymer means the actual measured value, actual value, or determined characteristic information of the characteristic information of the polymer.

[0088] (2) Construction of artificial neural network and input of learning data (T20)

[0089] Next, an artificial neural network model to be trained is constructed using the training data. As the artificial neural network model in the present invention, a known artificial neural network model is used, and in particular, an artificial neural network model of MPNN (message passing neural network) or GNN (graph neural network) based on message passing can be applied.

[0090] For example, when applying GNN, in the case of supervised learning, molecular learning data consisting of a set of {polymer graph information, predetermined characteristic information of the polymer} constructed above is input to a graph neural network (GNN) model. The predetermined characteristic information of the polymer includes, for example, the refractive index, glass transition temperature, density, etc. of the polymer.

[0091] At this time, the polymer graph information is an input value of the neural network model, and the specific property information of the polymer is input as a true value (labeled data) of the output value. At this time, the input polymer graph information includes at least one of the above-mentioned polymer graph node connection information, polymer graph node attribute information, and polymer graph edge attribute information, and can be input by being embedded in a polymer adjacency matrix and a polymer node / edge property matrix.

[0092] (3) Artificial Neural Network Model Training Step (T30)

[0093] The neural network model is trained by updating the parameters of the neural network model based on the input polymer graph information and the predetermined characteristic information of the polymer. The neural network training process conforms to a known neural network model training process, and as an example, the parameters of the artificial neural network are updated so that an error function defined as the difference between a predicted value of the characteristic information of the polymer calculated from the neural network model upon receiving the polymer graph information and the predetermined characteristic information of the polymer in the training data is minimized.

[0094] 3.2.Methods for analyzing polymer properties

[0095] This is a method for predicting the characteristic information of a selected polymer by inputting the graph information of the selected polymer to the artificial neural network generated according to the above-mentioned method for generating a polymer characteristic analysis model of the present invention. The method will be described with reference to Fig. 5(b).

[0096] (1) Polymer graph information pre-processing process (S10)

[0097] A preprocessing process is performed to obtain polymer graph information from the chemical structure of a selected polymer for which characteristic information is to be predicted. As described above, the polymer graph information includes information on the connection nodes and connection edges 40 between repeated monomers in addition to node and edge information on monomers constituting the polymer. That is, the polymer graph information is preprocessed to include at least one of the above-mentioned polymer graph node connection information, polymer graph node attribute information, and polymer graph edge attribute information.

[0098] (2) Polymer graph information input step (S20)

[0099] This is a step of inputting polymer graph information based on the chemical structure of the selected polymer into the previously trained neural network model. At this time, the input polymer graph information includes at least one of the above-mentioned polymer graph node connection information, polymer graph node attribute information, and polymer graph edge attribute information, and can be input by being embedded in the above-mentioned polymer adjacency matrix and polymer node / edge property matrix.

[0100] (3) Polymer property information prediction step (S30)

[0101] This is a step of calculating and outputting characteristic information of a selected polymer from the input polymer graph information of the selected polymer using the trained artificial neural network model.

[0102] 4. Polymer property information prediction system 300 according to the present invention

[0103] The polymer characteristic information prediction system according to the present invention will be described with reference to FIG.

[0104] (1) Polymer graph information input unit 100

[0105] The polymer graph data input unit 100 is a component that converts the chemical structure of a selected polymer into polymer graph information that describes the chemical structure of the selected polymer using graphical data and receives the input.

[0106] The polymer graph input unit 100 includes a polymer graph information acquisition unit 110 that acquires polymer graph information from the chemical structure of a selected polymer whose characteristic information is to be predicted.

[0107] The polymer graph information acquisition unit 110 generates polymer graph information by assigning connection node and connection edge information between repeated repeat units to respective variables in addition to node and edge information of monomers constituting the polymer as described above, and processing the information. That is, the polymer graph information is a data set including at least one of the above-mentioned polymer graph node connection information, polymer graph node attribute information, and polymer graph edge attribute information, and the polymer graph connection information is information representing a connection relationship between nodes corresponding to each atom constituting the repeat units forming the polymer and connection nodes connected to connection points, which are points where the repeat units are repeatedly connected, among the nodes.

[0108] The polymer graph input unit 100 may further include a matrix conversion unit 120 which, when inputting polymer graph information to the artificial neural network model unit 200 described later, generates and inputs a polymer adjacency matrix representing the mutual connection relationships between two or more nodes representing the respective atoms constituting a single molecule of a polymer and the connection relationship information between the connection nodes and the nodes, one or more edges representing bonds between the respective nodes, and a polymer edge attribute matrix including attribute information of the connection edges representing the connections between the nodes and the connection nodes.

[0109] In this case, as described above, the polymer graph information may include at least one of the following: polymer graph node connection information, which is information representing the mutual connection relationship between nodes corresponding to each atom constituting the repeating unit forming the polymer and connection nodes connected to connection points, which are points where the repeating units are repeatedly connected among the nodes; polymer node attribute information, which is information representing the attribute of nodes corresponding to each atom constituting the repeating unit forming the polymer and connection nodes connected to connection points, which are points where the repeating units are repeatedly connected among the nodes; and polymer edge attribute information, which is information representing the attribute of edges connecting nodes corresponding to each atom constituting the repeating unit forming the polymer and connection edges connecting at least one of the nodes to the connection node.

[0110] (2) Polymer Property Information Prediction Unit 200

[0111] The polymer characteristic information prediction unit 200 includes an artificial neural network 210 that is trained to predict and calculate predetermined characteristic information of the polymer from the polymer graph information.

[0112] The artificial neural network is configured to receive the above-mentioned polymer graph information as input and output a predicted value of predetermined characteristic information of the polymer, and is machine-learned by calculating a predicted value of predetermined characteristic information of a predetermined training polymer from the polymer graph information of the polymer, and updating the parameters of the neural network according to the above-mentioned procedure so as to minimize an error function defined as the difference from the label value of the characteristic information.

[0113] (3) Computer equipment

[0114] The polymer characteristic information prediction system 300 according to the present invention may be configured as a computer device having a single calculation system or multiple calculation systems connected via a network. The computer device may be configured with the above-mentioned polymer graph information input unit 100 and polymer characteristic information prediction unit 200 from a memory device 310 and a calculation device 320 that constitute the computer device. That is, the computer device 300 includes the memory device 310 and the calculation device 320, and the above-mentioned polymer graph information input unit 100 and polymer characteristic information prediction unit 200 occupy certain functions and parts of the memory device 310 and the calculation device 320.

[0115] The polymer graph information input unit 100 and the polymer characteristic information prediction unit 200 may include computer algorithms that are stored in the memory device 310 of the computer device and read from the arithmetic device 320 to perform each procedure when generating and processing the polymer graph information, or may be composed of the computer algorithms.

[0116] The reference numbers and names of the drawings used in the description of the present invention are as follows. [Explanation of symbols]

[0117] 10…node 20…Edge 30…Connection point 40…Connecting edge 100...Polymer graph information input section 110...polymer graph information acquisition unit 120...Matrix transformation section 200…Polymer Property Information Prediction Department 210…Artificial Neural Networks 310...Memory device 320...Arithmetic device

Claims

1. A method for converting a chemical structure of a polymer into computer-processable data, comprising the steps of: a polymer node connection variable setting step of setting, as a polymer node connection variable, a connection relationship between nodes representing each atom constituting the repeating unit of the selected polymer and a connection node connected to a connection point, which is a point where the repeating unit is repeatedly connected, among the nodes; a polymer graph node connection information assignment step of assigning attribute values ​​of mutual connection relationships between the nodes and the connection nodes to the set polymer node connection variables; A data processing method for polymer chemical structures, comprising:

2. a polymer node attribute variable setting step of setting attribute information of a node corresponding to each atom constituting a repeating unit forming the selected polymer and information of a connection node connected to a connection point where the repeating unit is repeatedly connected among the nodes as a polymer node attribute variable; a polymer graph node attribute information assignment step of assigning attribute values ​​of the nodes and the connection nodes to the set polymer node attribute variables; The data processing method for polymer chemical structures according to claim 1 , further comprising:

3. a polymer edge attribute variable setting step of setting information on edges connecting nodes corresponding to each atom constituting a repeating unit forming the selected polymer and a connection edge connecting at least one of the nodes and the connection node as polymer edge attribute variables; a polymer graph edge attribute information assignment step of assigning attribute values ​​of the edges and the connecting edges to the set polymer edge attribute variables; The data processing method for polymer chemical structures according to claim 2, further comprising:

4. 1. A method for generating a computer-implemented model that graphically describes the chemical structure of a selected polymer and analyzes selected properties of said polymer, comprising the steps of: A basic data acquisition step consisting of chemical structure information of a large number of learning polymers and known values ​​of predetermined characteristic information of the polymers; a data pre-processing step of converting chemical structure information of the learning polymer among the basic data into polymer graph information; a polymer characteristic information prediction artificial neural network configuration step for configuring a polymer characteristic information prediction artificial neural network so as to receive polymer graph information and output a predetermined characteristic information prediction value of the polymer; a learning data input step of inputting polymer graph information of the learning polymer and predetermined property information of the learning polymer into the polymer property information prediction artificial neural network; a learning step of the polymer characteristic information prediction artificial neural network, which updates parameters of the polymer characteristic information prediction artificial neural network based on a comparison result between a predicted value of a predetermined characteristic information of the learning polymer output by the polymer characteristic information prediction artificial neural network and a known value of the predetermined characteristic information of the learning polymer; A method for generating a polymer property analysis model, comprising:

5. The polymer graph information is Polymer graph node connection information is information representing a mutual connection relationship between nodes corresponding to each atom constituting the repeating unit forming the polymer and connection nodes connected to connection points, which are points where the repeating units are repeatedly connected, among the nodes; polymer node attribute information, which is information representing the attribute of a node corresponding to each atom constituting a repeating unit forming the polymer, and a connection node connected to a connection point, which is a point where the repeating unit is repeatedly connected, among the nodes; polymer edge attribute information representing attributes of edges connecting nodes corresponding to each atom constituting a repeating unit forming the polymer and a connection edge connecting at least one of the nodes to the connection node; The method for generating a polymer characteristic analysis model according to claim 4, comprising at least one of the following steps:

6. 1. A computer-implemented method for graphically describing the chemical structure of a selected polymer and analyzing selected properties of said polymer, comprising: generating a machine-trained artificial neural network that predicts a predetermined property of the selected polymer based at least in part on chemical structure data associated with the selected polymer; a polymer graph information acquisition step of acquiring polymer graph information that describes the chemical structure of the selected polymer by graphical data; a polymer graph input step of providing the obtained polymer graph information as an input of the machine-learned graph neural network; receiving prediction data describing the predetermined property of the selected polymer as an output of the machine-learned graph neural network; comprising Polymer graph information describing the chemical structure of the selected polymer by graph information, Polymer graph node connection information is information representing a mutual connection relationship between nodes corresponding to each atom constituting the repeating unit forming the polymer and connection nodes connected to connection points, which are points where the repeating units are repeatedly connected, among the nodes; polymer node attribute information, which is information representing the attribute of a node corresponding to each atom constituting a repeating unit forming the polymer, and a connection node connected to a connection point, which is a point where the repeating unit is repeatedly connected, among the nodes; polymer edge attribute information representing attributes of edges connecting nodes corresponding to each atom constituting a repeating unit forming the polymer and a connection edge connecting at least one of the nodes to the connection node; A computer-implemented method comprising at least one of the following:

7. Providing the obtained polymer graph information as an input of the machine-learned graph neural network includes: a polymer adjacency matrix constructing step of constructing an adjacency matrix representing a mutual connection relationship between two or more nodes representing each atom constituting the single molecule; a polymer adjacency matrix input step of inputting the constructed graph matrix as the graph neural network; Including, The polymer adjacency matrix is 7. The computer-implemented method of claim 6, further comprising connection relationship information between the connecting nodes and the nodes.

8. Providing the obtained polymer graph information as an input of the machine-learned graph neural network includes: a polymer edge attribute matrix input step of generating and inputting a polymer edge attribute matrix including attribute information of one or more edges representing connections between the respective nodes and connection edges representing connections between the nodes and the connection nodes; 8. The computer-implemented method of claim 7, comprising:

9. Obtaining, by one or more computing devices, a graph neural network trained by machine learning to predict a predetermined property of a polymer based at least in part on chemical structure data associated with the polymer, comprising: acquiring, by one or more computing devices, training data including a plurality of exemplary polymer chemical structures and predetermined property label values ​​describing predetermined properties of the exemplary polymers having the exemplary polymer chemical structures; a learning step of acquiring graph information that graphically describes a chemical structure of the exemplary polymer, inputting the graph information to a graph neural network, and training the graph neural network so that the graph neural network outputs a predetermined property label that describes a predetermined property of the exemplary polymer; 7. The computer-implemented method of claim 6, comprising:

10. The graph information graphically describing the chemical structure of the exemplary polymer is: Two or more pieces of node information representing the respective atoms constituting the single molecule forming the polymer; one or more edge information representing bonds between the respective atoms; Connection node information, which is a point where the single molecules are repeatedly connected; connection edge information representing connections between the atoms and the connection nodes; 10. The computer-implemented method of claim 9, comprising:

11. 1. A system for graphically describing the chemical structure of a selected polymer and analyzing selected properties of said polymer, comprising: a polymer graph information input unit for receiving input of polymer graph information by expressing a chemical structure of a selected polymer using the polymer graph information; a polymer characteristic information prediction unit that outputs a predicted value of predetermined characteristic information of the selected polymer from the polymer graph information; It is equipped with The polymer graph information is The polymer graph node connection information is information representing a connection relationship between nodes corresponding to each atom constituting the repeating unit forming the polymer and connection nodes connected to connection points, which are points where the repeating units are repeatedly connected, among the nodes; The polymer characteristic information prediction unit A polymer characteristic information prediction system comprising an artificial neural network trained to output a predetermined polymer characteristic information prediction value from polymer graph information including the polymer graph node connection information.

12. The polymer graph information input unit includes: Converting the chemical structure of the polymer into the polymer graph information; The polymer graph information is 12. The polymer characteristic information prediction system according to claim 11, further comprising at least one of: polymer node attribute information, which is information representing attributes of nodes corresponding to each atom constituting the repeating units forming the polymer, and connecting nodes connected to connection points among the nodes, which are points where the repeating units are repeatedly connected; and polymer edge attribute information, which is information representing attributes of edges connecting the nodes corresponding to each atom constituting the repeating units forming the polymer, and connecting edges connecting at least one of the nodes to the connecting node.

Images