Method and system for predicting glass transition temperature of polymer

Abstract

A computer-implemented method for estimating the glass transition temperature of a polymer from the chemical structure of the corresponding polymer is provided. The invention provides a method for graphically representing the chemical structure of a polymer. And a method and system for calculating a glass transition temperature of a corresponding polymer from graphic data of the polymer by machine learning a chemical structure of the polymer based on information defining an interconnection relationship between each atom constituting a repeating unit of the polymer and a connection node to which the repeating unit of the polymer is connected, wherein in addition to the chemical structure of the polymer, the molecular weight information is also reflected.

Description

Technical Field

[0001] The present invention relates to methods and systems for estimating the glass transition temperature of a polymer using an artificial intelligence neural network. Background Art

[0002] In order to input the structure of a compound into a computer device and perform a predetermined process, various methods for representing the structure of the compound as a structure recognizable by the computer device have been proposed. Representative methods include a method for representing the structure of a compound as a string using a simplified molecular linear input specification (SMILES), or a method for representing a string converted with SMILES as a descriptor.

[0003] However, it is difficult to represent polymers that are not single molecules in the SMILES method, so BigSMILES or hierarchical descriptors are proposed.

[0004] However, in such technology in the prior art, there are problems in that the larger the molecule, the more complicated the method for representing the molecule, the polymer is represented only as a fragment, so there is no technology for processing the polymer based on the representation, the properties of the polymer in which the repeating unit is infinitely repeated cannot be correctly represented, etc.

[0005] Among them, when estimating certain characteristics of a polymer, in the prior art, as shown in the following non-patent literature, by reflecting the structural characteristics of the polymer, certain characteristics of the polymer are estimated based on the reflected characteristics.

[0006] However, among the properties of polymers, since the glass transition temperature (Tg) is affected not only by the structure of the polymer but also by the molecular weight of the polymerized polymer, there is a problem that an accurate glass transition temperature cannot be estimated according to the technology in the related art.

[0007] The technology in the prior art is as follows.

[0008] Korean Patent Publication No. 2021-0042777 A

[0009] Korean Patent Publication No. 2021-0110539A

[0010] Machine-learning predictions of polymer properties with PolymerGenome,Journal of Applied Physics 128,171104,2020

[0011] Machine learning discovery of high-temperature polymers,Matter,Volume4,Issue 5,5May 2021,pages 1454-1456 Summary of the invention

[0012] Technical issues

[0013] Therefore, in order to solve the above problems, the present invention aims to provide a method for providing a data structure representing a polymer material as graphical data, and further, estimating the glass transition temperature of the polymer using the data structure representing the graphical data of the polymer, in which the molecular weight element of the polymer is reflected.

[0014] Technical Solutions

[0015] In order to solve the above problems, according to the present invention, there is provided a computer-implemented method for graphically describing the chemical structure of a selected polymer and calculating an estimated value of the glass transition temperature of the polymer, the above-mentioned computer-implemented method comprising: a polymer glass transition temperature characteristic information estimation artificial neural network acquisition step, which generates a machine-trained graphical artificial neural network for estimating glass transition temperature characteristic information about the selected polymer based at least in part on chemical structure data related to the selected polymer; a polymer graphic data acquisition step, which acquires polymer graphic data describing the chemical structure of the selected polymer as graphic data; a polymer graphic input step, which provides the acquired polymer graphic data as input to the graphic neural network; a polymer glass transition temperature characteristic information acquisition step, which receives polymer glass transition temperature characteristic information of the selected polymer from the output of the graphic neural network; and a polymer glass transition temperature calculation step, which calculates the glass transition temperature of the selected polymer by substituting molecular weight information about the selected polymer into the polymer glass transition temperature characteristic information.

[0016] In this case, the polymer graph data describing the chemical structure of the selected polymer as graph data may include at least one of the following: polymer graph node connection data, which is data representing the inter-attachment relationship between nodes corresponding to the respective atoms constituting the repeating units for constituting the polymer and connection nodes in the nodes connected to connection points, wherein the connection points are points at which the repeating units are repeatedly connected; polymer node attribute information, which is information representing attribute information about the nodes corresponding to the respective atoms constituting the repeating units for constituting the polymer and attributes of the connection nodes in the nodes connected to the connection points, wherein the connection points are points at which the repeating units are repeatedly connected; and polymer edge attribute information, which is information representing the attributes of the edges connecting the nodes corresponding to the respective atoms constituting the repeating units for constituting the polymer to each other and the attributes of the connection edges connecting at least one of the nodes to the connection nodes.

[0017] According to the present invention, there is also provided a glass transition temperature calculation system for a polymer, comprising: a data input unit, which receives polymer graphic data describing the chemical structure of a selected polymer as graphic data of the polymer and molecular weight data; an artificial neural network model unit, which estimates and calculates glass transition temperature characteristic information about the polymer based on the polymer graphic data; and a glass transition temperature calculation unit, which calculates the glass transition temperature of the polymer based on the glass transition temperature characteristic information about the polymer and the molecular weight data of the polymer, and in this case, the glass transition temperature characteristic information about the polymer is T in the following formula: g,inf The molecular weight data of the polymer is Mn in the following formula, and the glass transition temperature calculation unit calculates the glass transition temperature Tg according to the following formula.

[0018]

[0019] (Tg is the glass transition temperature of a polymer with a molecular weight Mn, T g,inf is the glass transition temperature when the molecular weight is infinite, K is a variable related to the free volume of the polymer, and Mn is the number average molecular weight of the polymer).

[0020] Beneficial Effects

[0021] According to the present invention, by providing a data structure capable of representing a polymer in computer-usable graphical data, a method and system for estimating characteristic information about a polymer material can be realized by machine learning the relationship between the molecular structure and characteristic information about the polymer material through a graph neural network (GNN) and a message passing neural network (MPNN). In addition, by reflecting the molecular weight factor when estimating the glass transition temperature of the polymer, a method and system for estimating the glass transition temperature of the polymer with improved accuracy can be provided compared to the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The accompanying drawings attached to the specification illustrate preferred embodiments of the present invention by way of example, and are used to enable the technical aspects of the present invention to be further understood together with the detailed description of the present invention given above, and therefore the present invention should not be interpreted solely by the contents in such drawings.

[0023] Figure 1 is a diagram showing a graphical representation of monomers and polymers according to the prior art.

[0024] Figure 2 is a diagram showing a graphical representation of a polymer according to the present invention.

[0025] Figure 3 is an example diagram for describing a graphical representation of a polymer according to the present invention.

[0026] Figure 4 is a diagram illustrating a process for implementing an artificial neural network for calculating the glass transition temperature of a polymer using a graphical representation of the polymer according to the present invention.

[0027] Figure 5 is a block diagram of a system for estimating the glass transition temperature of a polymer according to the present invention. DETAILED DESCRIPTION

[0028] 1. Graphical representation method of polymers according to the present invention

[0029] First, refer to Figure 1 and Figure 2 A method of graphically representing a polymer according to the present invention is described.

[0030] Figure 1 (a) and (b) in FIG. 1 are diagrams showing the structures of different polymers and monomers constituting each polymer. However, when trying to process molecular structure data using artificial neural networks, especially when representing molecular structures as graphs applied to GNNs, etc., there is no method for effectively representing the chemical structure of a polymer as a graph.

[0031] constitute Figure 1The monomers of the polymers shown in (a) and (b) are different from each other, but since they are the same except for some monomer ends, the graphics of the monomers are represented very similarly, so it is inaccurate to simply represent the graphics of the polymer as the graphics of the monomer and to represent the graphics of the polymer as a repetition of the graphics of the monomer.

[0032] Therefore, the present invention proposes Figure 2 A new method for representing polymer graphics as shown in FIG. Figure 2 A polymer graphics data method according to the present invention is described.

[0033] First, in the polymer graphic representation method of the present invention, monomers for constituting a polymer and repeating units repeated to constitute the polymer are derived. Figure 1 Examples of prior art representations of polymers and representations of monomers used to constitute the polymers are shown, and Figure 2 It is shown for the composition Figure 1 Diagram of the repeating unit of a polymer. Figure 2 In the repeating unit, for the polymer representation method of the present invention, each atom is represented as a node 10, and the coupling between atoms is represented as an edge 20. Next, when the repeating unit is repeated to constitute a polymer (which is one of the features of the present invention), the point where the repeating unit is repeatedly connected is represented as a connection point 30, and the node to which the connection point is connected is represented as a "connection node". In addition, the connection point 30 is a virtual representation of a connection node of adjacent repeating units connected by repetition, and the connection for connecting the connection point to the connection node is represented as a connection edge 40.

[0034] According to the polymer graph representation method of the present invention as described above, when an artificial neural network model is trained based on the chemical structure of a polymer, specifically, when learning is performed by applying artificial neural network models such as a message passing neural network (MPNN) and a graph neural network (GNN) based on message passing, learning can be performed by transmitting information about relative nodes connecting monomers even by adding a minimal amount of information other than monomer structure information, and learning can be performed by reflecting the unit repetition characteristics of the polymer.

[0035] 2. Computer-usable digitization of the graphical representation of the polymer according to the invention

[0036] The present invention converts the chemical structure of a polymer into a data structure that can be processed by a computer, i.e., "graphic data", according to the graphic representation method of the above-mentioned molecular or chemical structure of a polymer. In the present invention, this is referred to as "polymer graphic data".

[0037] According to the present invention, polymer graphic data that describes the chemical structure of a given or selected polymer in graphical data may include at least one of node data, edge data, connection node data, and connection edge data, the node data being data of a node 10 representing two or more nodes representing each atom of a repeating unit of a monomer used to constitute the polymer and data related to the node 10, the edge data being data related to one or more edges 20 representing couplings between the nodes, the connection node data being data related to at least one connection node among the nodes, the connection nodes being points at which the repeating units used to constitute the polymer are repeatedly connected, and the connection edge data being data related to the connection edges 40 that connect the connection points 30 to the connection nodes.

[0038] The difference between the monomer graph data of the present invention and the polymer graph data is that the polymer graph data of the present invention represents the nodes 10 connected to the connection points 30 and their edge information as connection nodes and connection edges in the graph data representation including the repeating units of the monomer graph data.

[0039] 2.1. Adjacency matrix and feature matrix

[0040] When converting the graph structure of a monomer into data that can be used on a computer, there may be various data representation methods, but generally, edge data and node data may be represented as an adjacency matrix and a feature matrix, respectively. Edge data is data related to coupling between atoms, and node data is data related to each atom.

[0041] Based on the polymer representation method according to the present invention, in addition to the monomer graph representation method in the related art, it is possible to convert the polymer graph into computer-recognizable data by additionally adding data of nodes and connection edges 40 connected to the connection points 30 .

[0042] (1) Adjacency Matrix Expression of Monomer Graph

[0043] For simplicity of description, the following is described by way of example. Figure 3 For the simple structure shown, the adjacency matrix A of the repeating unit of a monomer consisting of nodes 1, 2, 3 and 4 can be represented as follows.

[0044] For simplicity of description, the following is described by way of example. Figure 3 For the simple structure shown, the adjacency matrix A of the repeating unit of a monomer consisting of nodes 1, 2, 3 and 4 can be represented as follows.

[0045] Monomer adjacency matrix

[0046] The adjacency matrix A has as many rows and columns as the number of nodes, and each element Ai,j is data indicating whether the i-th (i=1, ..., 4) node and the j-th (j=1, ..., 4) node are connected to each other. The value of each matrix element represents data about the connection to other nodes. For example, since node 1 is only connected to node 2 and not to nodes 1, 3, and 4, A representing the adjacency matrix 1,j (j=1,...,4)=[0 1 0 0].

[0047] (2) Adjacency matrix representation of polymer graphs according to the present invention

[0048] According to the polymer graphic data representation method of the present invention, as a Figure 3 An example of an adjacency matrix representation of a polymer graph The polymer adjacency matrix PA can be represented as follows.

[0049] Polymer adjacency matrix

[0050] like Figure 3 As shown, the polymer adjacency matrix PA adds two connection points (5, 6), and these points refer to node 4 and node 1 of the adjacent monomers, respectively, so node 1 and node 4 are represented as being connected to node 4 and node 1, respectively. That is, since there are no nodes connected to node 1 and node 4 in the monomer, the matrix value indicating 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, since node 1 and node 4 are connected to node 4 and node 1 of the adjacent monomer, respectively, the polymer adjacency matrix PA is represented as PA 1,4 =1,PA 4,1 =1, and contains repeated connection data for the polymer. That is, since node 1 and node 4, which are not connected in the monomer, are connected to each other in the polymer, in the polymer adjacency matrix showing connection edge data, node 1 and node 4 are represented as connected to each other.

[0051] However, in this case, since node 1 and node 4 are not connected to each other within the monomer, in order to indicate that the connection is not a connection within the monomer but a node connected to an adjacent repeating unit, it can be distinguished by expressing the node connection relationship as a negative number to represent the polymer adjacency matrix as follows, assigning the characteristics of the "connected node connected to the connection point" to the characteristics of the corresponding node in the node feature matrix to be described below, or assigning the adjacent repeating unit connection attributes to the edge feature matrix.

[0052] Polymer adjacency matrix

[0053] In other words, in addition to the graphical data of the repeating units of the monomer, the graphical data representing the chemical structure of the polymer as a graph and describing it as graphical data according to the present invention also includes connection node data representing the connection nodes as nodes to which the monomer is repeatedly connected and data of the connection edges 40 connected to the connection nodes.

[0054] The expression method of the polymer adjacency matrix PA indicates that the nodes 1 and 4 of the repeating unit of the monomer composed of nodes 1, 2, 3 and 4 are connected to the nodes 4 and 1 of the adjacent repeating unit, respectively, and this connection is repeated. However, at this time, Figure 3 Nodes 1 and 4 in are not connected to each other within the monomer, but are connected to nodes of adjacent repeating units, so that these nodes can be distinguished by designating the nodes as "connecting nodes" and assigning connecting node attributes to the node feature matrix or assigning adjacent repeating unit connection attributes to the edge feature matrix.

[0055] This can be used as an embedding process for converting the chemical structure of a polymer into data that can be processed by a computer, which is achieved through a polymer edge variable setting step and a polymer graph edge data assignment step, the polymer edge variable setting step sets the interconnection relationship between the nodes representing the individual atoms constituting the monomer repeating unit of the selected polymer and the connecting nodes as the points where the monomers are repeatedly connected as polymer connection edge variables, and the polymer graph edge data assignment step assigns the attribute value of the interconnection relationship between the node and the connecting node to the set polymer edge variable. As shown in the polymer adjacency matrix PA above, the attribute value of the interconnection relationship between the node and the connecting node can be determined as "1" and "0", indicating connection / non-connection, and its matrix expression is a polymer adjacency matrix. The polymer adjacency matrix of the present invention has the same component values ​​as a typical adjacency matrix, but as described above, it differs from the typical adjacency matrix by adding connecting nodes to typical single-molecule graph nodes.

[0056] (3) Characteristic matrix of a single figure

[0057] In addition to the adjacency matrix showing the connection relationship between nodes, the graph representation of the compound may have a feature matrix representing predetermined attribute information about each node and predetermined attribute information about each edge as graph data. Figure 3 In the exemplary monomer graph of , a monomer node feature matrix containing attribute value information (eg, three pieces of attribute information) about each atom can be represented as follows.

[0058] Single node feature matrix The node feature matrix NF can have as many rows as the number of nodes and as many columns as the number of types of eigenvalues ​​to be represented for each node, and can therefore represent a specific property to be represented for each atom. Each component has the jth eigenvalue of the i-th node. For descriptive purposes, the monomer feature matrix NF is described as an example with three arbitrary property values ​​for each node.

[0059] The information of the monomer graph can be represented as an edge feature matrix EF, where each row can represent each edge of the monomer, and the column data of each row can be represented as a predetermined attribute value representing each edge. The edge attribute value may include the type of chemical bond (single bond, double bond, etc.), ring state, conjugated state, etc. For example, in Figure 3 In the case of a singleton in , since there are three connected edges, the edge feature matrix has three rows, and the edge attribute values ​​intended to contain information are assigned to the corresponding columns. An example of an edge feature matrix with three random edge attribute values ​​is as follows.

[0060] Monomer edge feature matrix

[0061] (4) Characteristic matrix of polymer graphics according to the present invention

[0062] The polymer graph data according to the present invention may have separate attribute values ​​added to the attribute values ​​of the above-mentioned monomer node feature matrix. For example, as Figure 3 The polymer node feature matrix PNF of the feature matrix of the exemplary polymer graph in may also have three columns plus attribute data, as in the example shown below, and the added columns are represented as attribute values ​​indicating whether each node is a connecting node. For example, in addition to the monomer node feature matrix NF containing three pieces of attribute information about the first four nodes, the polymer node feature matrix PNF below may also have connecting node attribute information in the fourth column, as shown below. Figure 3 In the example of , since node 1 and node 4 are connection nodes connected to adjacent repeating units, in the fourth column, node 1 and node 4 are represented as having a value of “1”, and the other nodes are represented as having a value of “0”.

[0063] Polymer node feature matrix

[0064] This can be achieved by a polymer node variable setting step and a polymer graph node data assignment step, wherein each atom of the monomer constituting the selected polymer and the connection node as the point to which the monomer is repeatedly connected are set as polymer node variables, and the polymer graph node data assignment step assigns the connection node to be connected to the set polymer node variable, and assigns a node attribute value to each node. The node attribute value assigned to the node may include atomic number, hybridization (SP3, SP2, etc.), number of hydrogens, number of electrons, ring state, etc., and in particular, as an embodiment of the present invention, there may be a connection node attribute value, which is an attribute value of whether the node is connected to an adjacent repeating unit.

[0065] As another embodiment of the present invention, the connection node data can be represented by an edge feature matrix. For example, as with the above polymer node feature matrix, instead of using a node attribute value to indicate whether a node is a connection node, or indicating it together with a node attribute value, an attribute value indicating whether a node is a connection node is indicated in the edge feature matrix. Figure 3 In the example of (b) in FIG. 1 , it can be seen that the data of the fourth and fifth rows are added to contain information about the edges connected to the connection nodes 1 and 4, that is, Figure 3 The edges 5_1 and 4_6 in (b) are removed, and an attribute information column indicating whether the node is a connected node is added in the fourth column.

[0066] Polymer edge feature matrix

[0067] That is, according to the present invention, a method of representing the structure of a polymer as a graph and converting it into computer-usable data may be selected from one or more of the following methods.

[0068] ① In the adjacency matrix representing the connection relationship between each node of the repeating unit formed by at least one monomer, the component of the "connection node" connected to the adjacent repeating unit is designated as the "connection" attribute.

[0069] ② The “connection status” attribute field is assigned to the node attribute matrix representing the attributes of each node of the repeating unit, and “connection” is assigned to the “connection status” attribute field of the “connection node”.

[0070] ③ Add the edge connected to the connection node as a connection edge to the edge attribute matrix of the repeating unit, assign the "connection status" attribute field to the edge attribute, and assign "connection" to the corresponding field of the connection edge.

[0071] 2.2. Methods for converting the chemical structure of polymers into data that can be processed by computers

[0072] In the present invention, when training the artificial neural network model described below and estimating the properties of a polymer material using the trained artificial neural network, the chemical structure of the polymer material must be converted into data so that the polymer can be processed by a computer, and the conversion into data can be performed as follows according to the method of the present invention.

[0073] (1) Repeating unit information acquisition step

[0074] This is a step of acquiring information about a repeating unit, which is a unit of a polymer composed of a repeating monomer. In order to convert the connection node data in the repeating unit into data, which is one of the features of the present invention, information about the repeating unit used to constitute the polymer is acquired. This is a step of confirming whether the repeating unit repeated to constitute the polymer is repeated as a single monomer or as a combination of single monomers and describing the repeating unit in a graphical representation. The repeating unit may be such as Figure 2 The single monomer shown in (a) in the above example may be Figure 2 The monomers shown in (b) are connected to each other.

[0075] (2) Steps for assigning connection information of polymer graph nodes

[0076] In the present invention, a method for processing polymer graph data includes a step of setting as a variable the interconnection relationship between nodes representing individual atoms constituting a repeating unit of a polymer selected as an analysis target and connection nodes among the nodes connected to connection points, which are points where the repeating units are repeatedly connected.

[0077] The attribute values ​​of the interconnection relationship between the node and the connection node are assigned to the polymer node connection variables set in this way. In order to assign the attribute value of each connection relationship to the polymer node connection variable, the polymer adjacency matrix mentioned above can be used.

[0078] (3) Steps for assigning attribute information to polymer graph nodes

[0079] The method for processing polymer graph data of the present invention may also include a polymer node attribute variable setting step, which sets the attribute information about the nodes corresponding to the respective atoms constituting the repeating units used to constitute the selected polymer and the attribute information about the connecting nodes connected to the connection points in the nodes as polymer node attribute variables, wherein the connection points are the points where the repeating units are repeatedly connected.

[0080] The method additionally includes a polymer graph node attribute information assignment step of assigning attribute values ​​of the node and the connecting node to the polymer node attribute variable set in this manner. The node attribute information has an attribute value indicating whether the node is a connecting node, and as the node attribute information for the connecting node, a connection attribute value indicating that the node is connected to a connecting node of an adjacent repeating unit in the polymer is given, unlike a node that is not a connecting node.

[0081] (4) Steps for allocating polymer graph edge attribute information

[0082] The method for processing polymer graph data of the present invention may further include a polymer edge attribute variable setting step of setting information about edge interconnection nodes corresponding to respective atoms constituting a repeating unit for constituting the selected polymer and a connecting edge connecting at least one of the nodes to a connecting node as a polymer edge attribute variable. A connecting edge is an edge connecting a connecting node to another node.

[0083] The present invention also includes a polymer graph edge attribute information assignment step, which assigns the attribute values ​​of the edges and connecting edges to the set polymer edge attribute variables, and assigns the connecting edge attribute value to the connecting edge (which means that the edge connects the connecting node to another node) as its edge attribute variable.

[0084] 3. A method according to the present invention for estimating the glass transition temperature of a polymer using a graphical representation of the polymer

[0085] Hereinafter, a method for processing polymer data in an artificial intelligence neural network according to the polymer graphical representation method of the present invention described above will be described.

[0086] The present invention provides a method for constructing an artificial neural network model for estimating the glass transition temperature of a polymer by applying the above-mentioned polymer graphical representation method and estimating the glass transition temperature of the polymer through machine learning.

[0087] 3.1. Method for generating polymer glass transition temperature estimation model

[0088] Figure 4 (a) in the figure shows a method for generating a polymer glass transition temperature estimation model according to the present invention. Based on this, a method for generating a computer-implemented polymer glass transition temperature estimation model will be described, which is used to estimate the glass transition temperature of a selected polymer according to the chemical structure of the corresponding polymer according to the graphical data representation method of the present invention.

[0089] The polymer glass transition temperature estimation model of the present invention is configured to estimate the glass transition temperature by applying the relationship between the glass transition temperature and the molecular weight of the polymer defined by the Flory-Fox equation of [Formula 1]. The glass transition temperature Tg of a polymer is a physical property value affected by both molecular structure and molecular weight, but the technology in the prior art fails to reflect the influence of both molecular structure and molecular weight when estimating the glass transition temperature. The present invention realizes an estimation model with improved accuracy by using the following formula to reflect both molecular structure and molecular weight when estimating the glass transition temperature. In addition, it has become possible to estimate Tg based on the molecular weight change of polymers of the same molecular structure.

[0090] [Formula 1]

[0091]

[0092] T g,inf is the glass transition temperature when the molecular weight is theoretically infinite, K is an experimental parameter related to the free volume in the polymer, and Mn is the number average molecular weight of the polymer.

[0093] Therefore, the estimation model of the present invention includes an artificial neural network that estimates T in Formula 1 based on the molecular structure of the selected polymer. g,inf and K, and is configured to be output by the artificial neural network T g,inf The and K values ​​and the molecular weight of the selected polymer are applied to Formula 1 to calculate the glass transition temperature Tg of the selected polymer. In this estimation model, the following process is performed to generate an estimated T g,inf and K values, which are characteristic information about the glass transition temperature of the polymer.

[0094] (1) Training data preparation step (T10)

[0095] In order to generate the above polymer glass transition temperature estimation model, training data for training an artificial neural network is prepared. The training data of the present invention includes a data set of {polymer graph data including structural information about sample polymers, measured Tg values ​​of sample polymers, and molecular weight Mn of sample polymers}.

[0096] In addition, the training data of the present invention includes a predetermined number of {polymer graph data containing structural information on the sample polymer, the measured Tg value of the sample polymer, and the molecular weight Mn of the sample polymer} or more data sets as a data set of polymers having similar structures but different molecular weights and a data set of polymers having different structures but the same molecular weight. By constructing the training data set in this way, the artificial neural network of the present invention to be described below can be trained so that the estimated Tg value (Tg estimated value) reflects both the structural characteristics and the molecular weight characteristics of the polymer.

[0097] (2) Artificial neural network model construction and learning data input steps (T20)

[0098] Next, an artificial neural network model to be trained using training data is constructed. The artificial neural network in the present invention uses a known artificial neural network, and in particular, an artificial neural network such as a message passing neural network (MPNN) and a graph neural network (GNN) based on message passing can be adopted.

[0099] For example, when GNN is used as an artificial neural network, the case of supervised learning is described, and the polymer training data of the data set containing the above-constructed {polymer graphic data containing structural information about the sample polymer, the measured Tg value of the sample polymer, and the molecular weight Mn of the sample polymer} is input into the GNN.

[0100] In this case, polymer graph data is input as input values ​​of the artificial neural network, and the known measured Tg value and molecular weight Mn of the corresponding polymer are input as true values ​​(labeled data) of output values ​​of the neural network. In this case, the input polymer graph data may include at least one of the polymer graph node connection information, polymer graph node attribute information, and polymer graph edge attribute information described above, and may be input by being embedded in a polymer adjacency matrix and a polymer node / edge feature matrix.

[0101] (3) Artificial neural network training steps (T30)

[0102] This is a process of training an artificial neural network by updating the neural network parameters of the artificial neural network based on the input polymer graph data, molecular weight information, and glass transition temperature characteristic information about the polymer. The learning process of the artificial neural network follows the known artificial neural network learning process, and as an example, the Tg estimate is calculated by receiving the polymer graph data of the training data and reflecting the measured molecular weight information in the training data set to the estimated value of the glass transition temperature characteristic information about the polymer calculated based on the artificial neural network according to [Formula 1], and the neural network parameters in the artificial neural network are updated so that the error function defined as the difference between the Tg estimate of the corresponding polymer in the training data and the measured Tg value is minimized. By updating the neural network parameters, the artificial neural network that calculates the glass transition temperature characteristic information based on the structure of the polymer is completed.

[0103] Therefore, in the present invention, accurate Tg estimation in polymers is possible because the artificial neural network is trained to minimize the error between the measured value and the Tg estimate that reflects the molecular weight information.

[0104] Among them, in the present invention, the term "neural network parameters" of the artificial neural network includes weights and biases adjusted when the artificial neural network model is trained according to training data, and adjusting such parameters makes the neural network model estimate more accurate outputs according to the input data, and is used in the general sense of the term in this field to indicate that it is optimized during the machine learning process to estimate the optimal output.

[0105] 3.2. Estimation method of polymer glass transition temperature

[0106] A process for estimating the glass transition temperature of a selected polymer by inputting graph data of the selected polymer into an artificial neural network generated according to the polymer glass transition temperature model generation method of the present invention and reflecting the number average molecular weight Mn of the selected polymer will be described.

[0107] (1) Polymer Graphic Data Preprocessing Process (S10)

[0108] like Figure 4 As shown in (b) of FIG. 1 , a preprocessing process for acquiring polymer graph data from the chemical structure of the selected polymer whose glass transition temperature is to be estimated is performed. As described above, in addition to the node and edge data of the monomers constituting the polymer, the polymer graph data also includes data on the connection edges 40 and connection nodes between the repeating monomers. That is, the polymer graph data 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.

[0109] (2) Polymer Graphic Data Input Step (S20)

[0110] This is a step of inputting polymer graph data based on the chemical structure of the selected polymer into the trained learning neural network model. In this case, the input polymer graph data may include at least one of polymer graph node connection information, polymer graph node attribute information, and polymer graph edge attribute information, and may be input by embedding in the above-mentioned polymer adjacency matrix and polymer node / edge feature matrix.

[0111] (3) Polymer Glass Transition Temperature Characteristic Information Estimation Step (S30)

[0112] This is a trained artificial neural network model that calculates and outputs the glass transition temperature characteristic information T of the selected polymer based on the input polymer graph data of the selected polymer. g,inf and K value steps.

[0113] (4) Molecular weight information reflecting step (S40)

[0114] This is based on the glass transition temperature characteristic information T calculated by the artificial neural network model. g,inf and K value, and the step of calculating the glass transition temperature Tg using the above [Formula 1]. The molecular weight of the polymer is obtained and reflected by a known method.

[0115] 4. A system for calculating the glass transition temperature of a polymer using a graphical representation of the polymer according to the present invention

[0116] (1) Data input unit 100

[0117] The data input unit 100 is configured to receive polymer graphic data describing the chemical structure of the selected polymer as graphic data and molecular weight data of the polymer, and further includes a data preprocessing unit (not shown) for converting the chemical structure of the polymer into polymer graphic data and inputting the converted data into the data input unit. The molecular weight data (molecular weight information) can be obtained by a known method.

[0118] When the polymer graph data is input into the artificial neural network model unit, the data preprocessing unit can generate and input a polymer adjacency matrix and a polymer edge attribute matrix, wherein the polymer adjacency matrix expresses the interconnection relationship between two or more nodes representing the atoms constituting a single molecule and the connection relationship information between connecting nodes, and the polymer edge attribute matrix includes attribute information about one or more edges representing the coupling between the nodes and the connecting edges representing the connection between the nodes and the connecting nodes.

[0119] In this case, as described above, the polymer graph data may include at least one of polymer graph node connection data, polymer node attribute information, and polymer edge attribute information, wherein the polymer graph node connection data is data representing the interconnection relationship between nodes corresponding to the respective atoms constituting the repeating units for constituting the polymer and connection nodes in the nodes connected to connection points, which are points at which the repeating units are repeatedly connected; the polymer node attribute information is data representing attribute information about the nodes corresponding to the respective atoms constituting the repeating units for constituting the polymer and attributes of connection nodes in the nodes connected to connection points, which are points at which the repeating units are repeatedly connected; and the polymer edge attribute information is information representing attributes of edges connecting nodes corresponding to the respective atoms constituting the repeating units for constituting the polymer to each other and of connection edges connecting at least one of the nodes to the connection nodes.

[0120] (2) Artificial neural network model unit 200

[0121] This is a configuration for estimating and calculating the glass transition temperature characteristic information about the polymer from the polymer graph data. The glass transition temperature characteristic information about the polymer corresponds to T in the above formula 1. g,inf and K.

[0122] Such an artificial neural network model unit is configured to receive polymer graph data and output an estimated value of glass transition temperature characteristic information of a corresponding polymer, and includes a polymer glass transition temperature characteristic information estimation artificial neural network, which is machine trained by calculating an estimated value of glass transition temperature characteristic information of a learning polymer based on polymer graph data of a predetermined learning polymer, calculating an estimated Tg value based on the calculated estimated value of glass transition temperature characteristic information of the learning polymer and measured molecular weight information about the learning polymer, and updating its neural network parameters so as to minimize an error function defined as the difference between the estimated Tg value and the measured Tg value of the polymer in the training data.

[0123] (3) Glass transition temperature calculation unit 300

[0124] The glass transition temperature calculation unit is configured to calculate the glass transition temperature of the polymer according to the above [Formula 1] based on the polymer glass transition temperature characteristic information calculated by the artificial neural network model unit and the molecular weight data of the polymer input from the data input unit.

[0125] The above has described a graphical data representation method for graphically representing the structure of a polymer according to the present invention, as well as a method and system for ultimately calculating the accurate glass transition temperature of a polymer by estimating glass transition temperature characteristic information about the polymer based on the graphical data of the polymer and reflecting the molecular weight information of the polymer in the estimated glass transition temperature characteristic information about the polymer.

[0126] With the method and system for calculating the glass transition temperature of a polymer reflecting molecular weight information according to the present invention, estimation adjustment can be made according to the molecular weight by considering the influence of the molecular weight alone, which has the effect of making it possible to construct a model capable of making accurate estimations.

[0127] For example, according to the model in the prior art, when the actual Tg of each of the samples of three types of polymers A, B and C having similar structures and molecular weights of 2000, 50000 and 80000 respectively is similar, since only the structure of the molecules is reflected without considering the molecular weight, learning and estimation are performed on the samples with similar structures so that the estimated Tg values ​​are similar regardless of the molecular weight.

[0128] That is, when the actual Tg (glass transition temperature) of polymer A having a molecular weight of K is R_Tg_A(K) and its estimated Tg is P_Tg_A(K), for R_Tg_A(2000)≈R_Tg_B(50000)≈R_Tg_C(80000), the prior art trains a neural network and performs an estimation as P_Tg_A(2000)≈P_Tg_B(50000)≈P_Tg_C(80000). In the case of the prior art, since the molecular weight is not reflected, even when samples A, B, and C have the same molecular weight of 50000, the prior art will estimate as P_Tg_A(50000)≈P_Tg_B(50000)≈P_Tg_C(50000). However, for A, B and C where R_Tg_A(2000)≈R_Tg_B(50000)≈R_Tg_C(80000), when compared under the same molecular weight, according to Formula 1, in fact, R_Tg_A(50000)>R_Tg_B(50000)>R_Tg_C(50000), so the technology in the prior art cannot be used to estimate the accurate glass transition temperature.

[0129] In contrast, in the present invention, when R_Tg_A(2000)≈R_Tg_B(50000)≈R_Tg_C(80000), even when the neural network has been trained such that P_Tg_A(2000)≈P_Tg_B(50000)≈P_Tg_C(80000), since the training of the neural network is performed by reflecting the molecular weight, the neural network parameters for adjusting the influence of the molecular structure are learned differently. Thus, for the same molecular weight, it is estimated that P_Tg_A(50000)>P_Tg_B(50000)>P_Tg_C(50000).

[0130] In another case, according to the prior art, when the molecular weights of polymers A, B, and C with similar structures are different and the actual Tg is different, i.e., when R_Tg_A(2000)<R_Tg_B(50000)<R_Tg_C(80000), since the training is performed based on the molecular structure and the actual Tg (which is the label value) is given differently, even when the polymers have similar structures, it is learned and estimated that P_Tg_A(2000)<P_Tg_B(50000)<P_Tg_C(80000). However, at this time, when the molecular weights of polymers A, B, and C are the same, according to [Equation 1], the actual Tg may be similar. In this case, the estimated values should be P_Tg_A(50000)≈P_Tg_B(50000)≈P_Tg_C(50000), but since the model according to the prior art is trained only based on the molecular structures of polymers A, B, and C with the same molecular weight, there is a problem in the case of estimating P_Tg_A(50000)<P_Tg_B(50000)<P_Tg_C(50000).

[0131] In contrast, according to the present invention, when R_Tg_A(2000)<R_Tg_B(50000)<R_Tg_C(80000), since the training is performed by reflecting the molecular weight, even when trained such that P_Tg_A(2000)<P_Tg_B(50000)<P_Tg_C(80000), for the same molecular weight, it is not estimated that P_Tg_A(50000)<P_Tg_B(50000)<P_Tg_C(50000), but rather P_Tg_A(50000)≈P_Tg_B(50000)≈P_Tg_C(50000).

[0132] As described above, according to the present invention, since the neural network model is trained to estimate the glass transition temperature by reflecting not only the molecular structure of the polymer but also the molecular weight information about the polymer, the influence of each of the molecular structure information and the molecular weight information can be learned, and thus the glass transition temperature can be estimated more accurately than the prior art.

[0133] The names of the respective reference numerals used in the present invention are as follows.

[0134] 10 nodes

[0135] 20 Edge

[0136] 30 connection points

[0137] 40 Connecting Edges

[0138] 100 data input units

[0139] 200 Artificial Neural Network Model Units

[0140] 300 Glass transition temperature calculation unit

Claims

1. A method for generating a computer-implemented model for estimating the glass transition temperature of a selected polymer, the method comprising: a training data acquisition step of constructing a training data set having structural information about a plurality of learning polymers, known measured values ​​of glass transition temperatures of the corresponding polymers, and measured molecular weights of the corresponding polymers; A polymer glass transition temperature characteristic information estimation artificial neural network construction step, which constructs a polymer glass transition temperature characteristic information estimation artificial neural network for receiving structural information about a polymer and outputting an estimated value of the glass transition temperature characteristic information about the polymer; a training data input step of inputting polymer structure information on the learning polymer and a measured value of the glass transition temperature of the learning polymer into the polymer glass transition temperature characteristic information estimation artificial neural network; as well as A training step of the polymer glass transition temperature characteristic information estimation artificial neural network, wherein a neural network parameter of the polymer glass transition temperature characteristic information estimation artificial neural network is updated based on a comparison result between an estimated value of the glass transition temperature calculated by reflecting the measured molecular weight of the corresponding polymer in the training data set in an estimated value of the glass transition temperature characteristic information of the learning polymer output by the polymer glass transition temperature characteristic information estimation artificial neural network and a measured value of the glass transition temperature of the learning polymer.

2. The method according to claim 1, wherein: The glass transition temperature characteristic information is the Tg,inf value and K value in the following formula 1; [Formula 1] (Tg is the glass transition temperature of a polymer with a molecular weight Mn, T g,inf is the glass transition temperature when the molecular weight is infinite, K is a variable related to the free volume of the polymer, and Mn is the number average molecular weight of the polymer).

3. The method according to claim 2, wherein: The polymer structure information is converted into polymer graphic data and input into the polymer glass transition temperature characteristic information estimation artificial neural network, and The polymer graphic data includes at least one of the following: Polymer graph node connection data, which is data indicating an interconnection relationship between nodes corresponding to respective atoms constituting a repeating unit constituting the polymer and connection nodes among the nodes connected to connection points, the connection points being points at which the repeating units are repeatedly connected; Polymer node attribute information, which is data indicating attribute information on nodes corresponding to respective atoms constituting repeating units constituting the polymer and attributes of connection nodes among the nodes connected to the connection points at which the repeating units are repeatedly connected; and The polymer edge attribute information is information indicating the attributes of the edge for connecting nodes corresponding to the respective atoms constituting the repeating unit constituting the polymer to each other and the connection edge for connecting at least one of the nodes to the connection node.

4. A computer-implemented method for calculating an estimate of the glass transition temperature of a selected polymer, the computer-implemented method comprising: a polymer glass transition temperature characteristic information estimation artificial neural network acquisition step, which generates a machine-trained graphical artificial neural network for estimating glass transition temperature characteristic information about a selected polymer based at least in part on a chemical structure associated with the selected polymer; a polymer graphic data acquisition step of acquiring polymer graphic data describing the chemical structure of the selected polymer as graphic data; a polymer graph input step, which provides the acquired polymer graph data as input to the graph artificial neural network; a polymer glass transition temperature characteristic information acquisition step, which receives the polymer glass transition temperature characteristic information of the selected polymer from the output of the graphical artificial neural network; and A polymer glass transition temperature calculation step of calculating the glass transition temperature of the selected polymer by substituting molecular weight information on the selected polymer into the polymer glass transition temperature characteristic information.

5. The computer-implemented method of claim 4, wherein: The polymer graphic data describing the chemical structure of the selected polymer as graphic data includes at least one of the following: Polymer graph node connection data, which is data indicating an interconnection relationship between nodes corresponding to respective atoms constituting a repeating unit constituting the polymer and connection nodes among the nodes connected to connection points, the connection points being points at which the repeating units are repeatedly connected; Polymer node attribute information, which is data indicating attribute information on nodes corresponding to respective atoms constituting repeating units constituting the polymer and attributes of connection nodes among the nodes connected to the connection points at which the repeating units are repeatedly connected; and The polymer edge attribute information is information indicating an attribute of an edge for connecting nodes corresponding to respective atoms constituting a repeating unit constituting the polymer to each other and a connecting edge for connecting at least one of the nodes to the connecting node.

6. The computer-implemented method of claim 5, wherein: The polymer graph input step of providing the acquired polymer graph data as input to the machine trained graph neural network comprises: a polymer adjacency matrix construction step, which constructs an adjacency matrix representing an interconnection relationship between two or more nodes, the two or more nodes representing individual atoms constituting the single molecule; and a polymer adjacency matrix input step, which inputs the constructed graph matrix into the graph artificial neural network, and The polymer adjacency matrix also includes connection relationship information between the connection nodes and the nodes.

7. The computer-implemented method of claim 4, wherein: The graphical artificial neural network is trained by: calculating an estimated value of glass transition temperature characteristic information about a learning polymer based on structural information about a plurality of polymers; and Neural network parameters are updated to minimize an error between an estimated value of a glass transition temperature calculated by reflecting a measured molecular weight of the corresponding polymer in the calculated estimated value of the glass transition temperature characteristic information and a measured value of the glass transition temperature of the learning polymer.

8. The computer-implemented method of claim 4, wherein: The step of obtaining the polymer glass transition temperature characteristic information estimation artificial neural network comprises: A step of acquiring, by one or more computing devices, training data, the training data comprising chemical structures of a plurality of exemplary polymers and measured values ​​of glass transition temperatures and molecular weights of the exemplary polymers having the chemical structures of the exemplary polymers; and A training step, which obtains graphical data that graphically describes the chemical structure of the exemplary polymer and inputs the graphical data into the graphical neural network, and trains the graphical neural network so as to output an estimated value of the glass transition temperature characteristic information of the exemplary polymer output by the graphical neural network.

9. The computer-implemented method of claim 8, wherein: Graphical data graphically describing the chemical structure of the exemplary polymer includes: Two or more pieces of node data representing individual atoms constituting a single molecule constituting the polymer; one or more edge data, which represent couplings between the atoms; connection node data, which are the points at which the individual molecules are repeatedly connected; and Connection edge data, which represents the connection between the atom and the connection node.

10. A computer-implemented method according to any one of claims 4 to 9, wherein: The glass transition temperature characteristic information is the Tg,inf value and K in the following formula 2; [Formula 2] (Tg is the glass transition temperature of a polymer with a molecular weight Mn, T g,inf is the glass transition temperature when the molecular weight is infinite, K is a variable related to the free volume of the polymer, and Mn is the number average molecular weight of the polymer).

11. A computer-implemented method according to any one of claims 4 to 9, wherein: The polymer glass transition temperature calculation step includes calculating the glass transition temperature of the polymer by substituting the number average molecular weight information of the polymer into the following formula 3; [Formula 3] (Tg is the glass transition temperature of a polymer with a molecular weight Mn, T g,inf is the glass transition temperature when the molecular weight is infinite, K is a variable related to the free volume of the polymer, and Mn is the number average molecular weight of the polymer).

12. A system for calculating the glass transition temperature of a polymer, comprising: a data input unit that receives structural information about a selected polymer and molecular weight data of the polymer; an artificial neural network model unit that calculates an estimate of glass transition temperature characteristic information about the polymer based on structural information about the polymer; and A glass transition temperature calculation unit calculates an estimated value of the glass transition temperature of the polymer based on the estimated value of the glass transition temperature characteristic information on the polymer and the molecular weight data of the polymer.

13. The glass transition temperature calculation system for polymers according to claim 12, wherein: The glass transition temperature characteristic information about the polymer is the Tg,inf value and K in the following formula 4, The molecular weight data of the polymer is Mn in the following formula 4, and The glass transition temperature calculation unit calculates the glass transition temperature Tg according to the following formula 4: [Formula 4] (Tg is the glass transition temperature of a polymer with a molecular weight Mn, T g,inf is the glass transition temperature when the molecular weight is infinite, K is a variable related to the free volume of the polymer, and Mn is the number average molecular weight of the polymer).

14. The glass transition temperature calculation system for polymers according to claim 13, further comprising a data preprocessing unit for converting structural information about the polymer into polymer graphic data, in, The artificial neural network model unit is machine trained to calculate an estimated value of glass transition temperature characteristic information about the polymer based on the polymer graphic data, and The polymer graphic data includes at least one of the following: Polymer graph node connection data, which is data indicating an interconnection relationship between nodes corresponding to respective atoms constituting a repeating unit constituting the polymer and connection nodes among the nodes connected to connection points, the connection points being points at which the repeating units are repeatedly connected; Polymer node attribute information, which is data indicating attribute information on nodes corresponding to respective atoms constituting repeating units constituting the polymer and attributes of connection nodes among the nodes connected to the connection points at which the repeating units are repeatedly connected; and The polymer edge attribute information is information indicating an attribute of an edge for connecting nodes corresponding to respective atoms constituting a repeating unit constituting the polymer to each other and a connecting edge for connecting at least one of the nodes to the connecting node.

15. The glass transition temperature calculation system for polymers according to claim 12, wherein: The artificial neural network model unit includes a polymer glass transition temperature characteristic information estimation artificial neural network, the polymer glass transition temperature characteristic information estimation artificial neural network is configured to receive structural information about the polymer and output an estimated value of the glass transition temperature characteristic information about the polymer, The polymer glass transition temperature characteristic information estimation artificial neural network is machine trained by: using structural information about a predetermined learning polymer and a measured value of the glass transition temperature and a measured value of the molecular weight of the learning polymer as training data, calculating an estimated value of the glass transition temperature characteristic information about the learning polymer based on the structural information about the learning polymer, calculating an estimated value of the glass transition temperature of the learning polymer using the estimated value of the glass transition temperature characteristic information and the measured value of the molecular weight, and updating the neural network parameters of the polymer glass transition temperature characteristic information estimation artificial neural network by comparing the estimated value of the glass transition temperature of the learning polymer with the measured value.

16. The glass transition temperature calculation system for polymers according to claim 15, wherein: The training data comprises a data set including: polymer graph data including structural information about a sample polymer, a measured Tg value of the sample polymer, and a molecular weight Mn of the sample polymer, and At least a predetermined number of the data sets are data sets of polymers having similar structures and different molecular weights and data sets of polymers having different structures and the same molecular weight.

Images