Method and system for predicting the glass transition temperature of polymers

Abstract

The present invention provides a computer-implemented method for predicting the glass transition temperature of a polymer from the chemical structure of the polymer. The present invention provides a method for graphically representing the chemical structure of a polymer, and machine-learns the chemical structure of the polymer based on information defining the connection relationship between each atom constituting the repeating unit of the polymer and the connection nodes to which the repeated repeating units are connected, and calculates the glass transition temperature of the polymer from the graph information of the polymer. In addition to the chemical structure of the polymer, the present invention provides a method and system for calculating the glass transition temperature of the polymer while also reflecting the molecular weight information.

Description

[Technical Field]

[0001] This invention relates to a method and system for predicting the glass transition temperature of polymers using an artificial intelligence neural network. [Background technology]

[0002] To input the structure of a compound into a computer and perform a predetermined process, a wide variety of methods have been proposed for representing the compound structure in a way that the computer can recognize. Typical examples include SMILES, which represents the compound structure as a string, or a method that represents the string converted to SMILES using a descriptor.

[0003] However, since polymers, which are not single molecules, are difficult to represent using the SMILES method, BigSMILES and hierarchical descriptors have been proposed.

[0004] However, conventional technologies have problems such as the representation of polymers becoming increasingly complex as the molecules get larger, and being limited to representing polymers in a fragmented manner. There is no technology that can process molecules based on such representations, and it is not possible to properly represent the properties of polymers in which repeating units are infinitely repeated.

[0005] On the other hand, in predicting the specific properties of a polymer, conventional methods have involved reflecting the structural properties of the polymer and predicting those properties from them, as disclosed in the non-patent literature below.

[0006] However, among the properties of polymers, the glass transition temperature (Tg) is influenced not only by the structure of the polymer but also by the molecular weight of the polymerized polymer. Therefore, conventional techniques have the disadvantage of not being able to accurately predict the glass transition temperature. [Prior art documents] [Patent Documents]

[0007] [Patent Document 1] Republic of Korea Patent Publication No. 2021-0042777 [Patent Document 2] Republic of Korea Patent Publication No. 2021-0110539 [Non-Patent Document]

[0008] [Non-Patent Document 1] Machine-learning predictions of polymer properties with Polymer Genome, Journal of Applied Physics 128, 171104, 2020 [Non-Patent Document 2] Machine learning discovery of high-temperature polymers, Matter, Volume 4, Issue 5, 5 May 2021, pages 1454-1456 [Summary of the Invention] [Problems to be Solved by the Invention]

[0009] Therefore, in order to solve the above problems, the present invention provides a data structure for representing a polymer substance by graphic information, and further predicts the glass transition temperature of a polymer using the data structure representing the graphic information of the polymer, and aims to provide a method and a system for predicting while reflecting the molecular weight element of the polymer. [Means for Solving the Problems]

[0010] To solve the above problems, the present invention provides a computer implementation method for graphically describing the chemical structure of a selected polymer and calculating a predicted glass transition temperature of the polymer, comprising: a polymer glass transition temperature characteristic information prediction artificial neural network acquisition step, which generates a graph artificial neural network that is machine-learned to predict glass transition temperature characteristic information of the polymer based at least partially on the selected polymer and associated chemical structure data; a polymer graph information acquisition step, which acquires polymer graph information that graphically describes the chemical structure of the selected polymer; a polymer graph input step, which provides the acquired polymer graph information as input to the graph 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 graph neural network; and a polymer glass transition temperature calculation step, which calculates the glass transition temperature of the selected polymer by substituting the molecular weight information of the selected polymer into the polymer glass transition temperature characteristic information.

[0011] In this case, the polymer graph information that describes the chemical structure of the selected polymer using graph information is characterized by including at least one of the following: polymer graph node connection information, which is information that represents the connection relationships between nodes corresponding to each atom constituting the repeating unit that forms the polymer and connection nodes that are connected to connection points, which are locations where the repeating unit is repeatedly connected; polymer node attribute information, which is information that represents the attribute of the node corresponding to each atom constituting the repeating unit that forms the polymer and the attribute of the connection node that is connected to connection points, which are locations where the repeating unit is repeatedly connected; and polymer edge attribute information, which is information that represents the attribute of the edges that connect the nodes corresponding to each atom constituting the repeating unit that forms the polymer and the connection edge that connects at least one of the nodes to the connection node.

[0012] Furthermore, the present invention provides a polymer glass transition temperature calculation system, which comprises: a data input unit that receives polymer graph information describing the chemical structure of a selected polymer using graphical data and molecular weight data of the polymer; an artificial neural network model unit that predicts and calculates the glass transition temperature characteristic information of the polymer from the polymer graph information; and a glass transition temperature calculation unit that calculates the glass transition temperature of the polymer from the glass transition temperature characteristic information and molecular weight data of the polymer. In this case, the glass transition temperature characteristic information of the polymer is T in the following equation 1. g,inf The values ​​are K, and the molecular weight data of the polymer is M in the following formula. n The glass transition temperature calculation unit calculates the glass transition temperature T from the following formula. g Calculate.

[0013]

number

[0014] (T g The molecular weight is M n The glass transition temperature of a polymer is T g,inf is the glass transition temperature when the molecular weight is infinite, K is the free volume relation variable of the polymer, and M n (This is the number-average molecular weight of the polymer.) [Effects of the Invention]

[0015] According to the present invention, by providing a data structure that allows polymers to be represented by computer-usable graphical information, it is possible to realize a method and system for predicting the properties of polymer materials by machine learning the relationship between the molecular structure and property information of polymer materials using a graph neural network (GNN) or a message-passing neural network (MPNN). Furthermore, by reflecting molecular weight elements when predicting the glass transition temperature of polymers, it is possible to provide a method and system for predicting the glass transition temperature of polymer property information with improved accuracy compared to conventional techniques.

[0016] The drawings accompanying this specification illustrate preferred embodiments of the present invention and are intended to further illustrate the technical idea of ​​the invention along with the content of the invention; therefore, the present invention shall not be construed as being limited only to what is shown in the drawings. [Brief explanation of the drawing]

[0017] [Figure 1] (a) and (b) are diagrams showing graphic representation methods of monomers and polymers in the conventional technology. [Figure 2] (a) and (b) are diagrams showing graphic representations of polymers according to the present invention. [Figure 3] (a) and (b) are illustrative diagrams for explaining the graphical representation of the polymer according to the present invention. [Figure 4] (a) and (b) are diagrams illustrating the procedure for realizing an artificial neural network that calculates the glass transition temperature of a polymer using a graphical representation of the polymer according to the present invention. [Figure 5] This is a block diagram of the polymer glass transition temperature prediction system according to the present invention. [Modes for carrying out the invention]

[0018] 1. Graph representation method of polymers according to the present invention

[0019] First, the method for graphically representing polymers according to the present invention will be explained based on Figures 1 and 2.

[0020] Figures 1(a) and 1(b) show the structures of different polymers and the monomers that make up each polymer. However, when attempting to process molecular structure data using artificial neural networks, especially when trying to represent molecular structures graphically for application to GNNs, there has been no effective way to represent the chemical structure of polymers graphically.

[0021] Although the monomers forming the polymers shown in Figures 1(a) and (b) are different from each other, they are identical except for a small portion of the monomer ends. Therefore, their monomer graphs appear very similar, and it was inaccurate to simply represent the polymer graph as a monomer graph and the polymer graph as a repetition of the monomer graph.

[0022] Therefore, in this invention, we propose a novel graph representation method for polymers as shown in Figure 2. The graph information representation method for polymers according to the present invention will be described below based on Figure 2.

[0023] First, in the polymer graph representation method of the present invention, monomers constituting the polymer and repeating units that repeatedly constitute the polymer are derived. Figure 1 shows an example of a conventional polymer representation and a representation of the monomers constituting the polymer, and Figure 2 shows the repeating units constituting the polymer in Figure 1. In the repeating units in Figure 2, for the polymer representation method of the present invention, each atom is represented by a node 10, and the bonds between the atoms are represented by edges 20. Next, in one of the features of the present invention, when the repeating units repeat to form a polymer, the points where the repeating units are repeatedly connected are represented by attachment points 30, and the nodes to which the attachment points are connected are represented by "attachment nodes". Furthermore, the attachment points 30 virtually represent the attachment nodes of adjacent repeating units that are repeatedly connected, and the connections between the attachment points and the attachment nodes are represented by attachment edges 40.

[0024] According to this graph representation method for polymers of the present invention, when learning an artificial neural network model based on the chemical structure of a polymer, particularly when applying a message-passing-based MPNN (message-passing neural network) or GNN (graph neural network) artificial neural network model, it becomes possible to learn by passing information from the opposite node to which the monomer is connected, even with only minimal additional information on top of the monomer structure information. This enables learning that reflects the unit repeatability characteristics of the polymer.

[0025] 2. Converting the graphical representation of the polymer according to the present invention into data that can be used on a computer.

[0026] This invention converts the chemical structure of a polymer into a data structure usable for computer processing, i.e., "graph data," by using a graph representation method that graphically represents the molecular or chemical structure of the polymer described above. In this invention, this is referred to as "polymer graph data."

[0027] The polymer graph information according to the present invention, which describes the chemical structure of a predetermined or selected polymer using graphic information, includes node data, which is data associated with two or more nodes representing each atom constituting a repeating unit of monomers making up the polymer and nodes 10; edge data, which is data associated with one or more edges 20 representing the bonds between each of the nodes; attaching node data, which is data associated with at least one attachment node, which is a location where the repeating units making up the polymer are repeatedly connected; and attaching edge data, which is data associated with an attachment edge 40, which is a connection connecting the attachment point 30 and the attachment node.

[0028] The difference between monomer graph information and polymer graph information of the present invention lies in the fact that, in the representation of graph information of repeating units including the monomer graph information, the polymer graph information of the present invention represents the node 10 connected to the connection point 30 and its edge information using connection nodes and connection edges.

[0029] 2.1. Adjacency matrix and feature matrix

[0030] When digitizing the graph structure of a monomer so that it can be used by a computer, there can be various data representation methods. Usually, the edge information and node information can be represented by an adjacency matrix and a characteristic matrix respectively. The edge information is information related to the bonds between atoms, and the node information is information related to each atom.

[0031] According to the polymer representation method of the present invention, in addition to the conventional monomer graph representation method, the information of the nodes and connection edges 40 connected to the connection point 30 can be further added to convert the polymer graph into data recognizable by a computer.

[0032] (1) Representation of the adjacency matrix of the monomer graph

[0033] For the sake of simplifying the explanation, taking a simple structure as shown in FIG. 3 as an example, the adjacency matrix A of the monomer repeating unit composed of nodes 1, 2, 3, and 4 can be represented as follows.

[0034]

Equation

[0035] The adjacency matrix A has rows and columns corresponding to the number of nodes. Each component A- i,j represents information indicating whether the i-th (i = 1,..., 4) node and the j-th (j = 1,..., 4) node are connected. The value of each matrix component indicates the connection information with other nodes. For example, since node 1 is only connected to node 2 and not connected to nodes 1, 3, and 4, A 1,j of the adjacency matrix is represented by [0 1 0 0] (j = 1,..., 4).

[0036] (2) Representation of the adjacency matrix of the polymer graph according to the present invention

[0037] According to the polymer graph information representation method of the present invention, the polymer adjacence matrix PA, which is a representation of the adjacency matrix of the polymer graph illustrated in Figure 3, can be represented as follows.

[0038]

number

[0039] As shown in Figure 3, the polymer adjacency matrix PA has two additional connection points 5 and 6, which represent nodes 4 and 1 of the adjacent monomers, respectively. Therefore, nodes 1 and 4 are represented as being connected to nodes 4 and 1, respectively. In other words, in the monomer, there are no nodes to which nodes 1 and 4 are connected, so the matrix value representing the connection relationship between node 1 and node 4 is A. 1,4 =0, A 4,1 Although it was =0, in the polymer representation method of the present invention, nodes 1 and 4 are connected to nodes 4 and 1 of adjacent monomers, respectively, so the PA of the polymer adjacency matrix PA 1,4 =1, PA 4,1 The value =1 indicates that it contains repeating connection information for the polymer. In other words, nodes 1 and 4, which were not connected in the monomer, are connected to each other in the polymer, and therefore, in the polymer adjacency matrix that displays the connection edge information, they are represented as being connected to each other.

[0040] However, in this case, since nodes 1 and 4 are not connected to each other within the monomer, to indicate that the connection is not within the monomer but between adjacent repeating unit nodes, the node connection relationship can be represented by a negative number and the polymer adjacency matrix can be represented as shown below, or a "connected node" characteristic that connects to the connection point can be assigned to the node's characteristics in the node characteristics matrix described later, or an adjacent repeating unit connection attribute can be assigned to the edge characteristics matrix to distinguish it.

[0041]

number

[0042] To reiterate, the graph information that graphically represents the chemical structure of the polymer according to the present invention and describes it using graphic data includes, in addition to graph information of monomer repeating units, connection node information representing connection nodes which are nodes to which monomers are repeatedly connected, and information of connection edges 40 connected to the connection nodes.

[0043] According to the representation method of the polymer adjacency matrix PA described above, a monomer repeating unit composed of nodes 1, 2, 3, and 4 is shown to be repeated by being connected at nodes 1 and 4 to nodes 4 and 1 of adjacent repeating units, respectively. However, in this case, nodes 1 and 4 in Figure 3 are not connected to each other within the monomer, but rather to the nodes of adjacent repeating units. Therefore, these nodes can be designated as "connecting nodes" and distinguished by assigning a connecting node attribute to the node characteristic matrix or an adjacent repeating unit connection attribute to the edge characteristic matrix.

[0044] This can be used as an embedding procedure to convert the chemical structure of a polymer into computer-processable data. This is achieved using a polymer edge variable setting step, which sets the connection relationships between nodes representing each atom constituting the monomer repeating unit of a selected polymer and connection nodes, which are the locations where the monomers are repeatedly connected, as polymer connection edge variables, and a polymer graph edge information assignment step, which assigns attribute values ​​of the connection relationships between the nodes and connection nodes to the set polymer edge variables. As is clear from the polymer adjacency matrix PA above, the attribute values ​​of the connection relationships between the nodes and connection nodes can be determined as "1" or "0" indicating connected / disconnected, and the polymer adjacency matrix is ​​a matrix representation of this. The polymer adjacency matrix of the present invention has component values, similar to a normal adjacency matrix, except that, as mentioned above, it differs because connection nodes are added to the normal monomolecule graph nodes.

[0045] (3) Characteristic matrix of monomer graph

[0046] In addition to the adjacency matrix that shows the connection relationships between nodes, the graph representation of a compound can also include a characteristic matrix as graph information that represents predetermined attribute information for each node and predetermined attribute information for each edge. In the exemplary monomer graph of Figure 3, the monomer node characteristic matrix that has attribute value information for each atom (exemplary, three pieces of attribute information) can be represented as follows.

[0047]

number

[0048] The node characteristic matrix NF has a number of rows corresponding to the number of nodes and a number of columns corresponding to the number of types of characteristic values ​​to be represented for each node, and can represent the predetermined attributes to be represented for each atom. i,j This has the j-th characteristic value of the i-th node. The monomer characteristic matrix NF is shown as an example where each node has three arbitrary attribute values ​​for illustrative purposes.

[0049] A monomer graph can be represented by an edge characteristic matrix EF, where each row represents an edge of the monomer, and the column data in each row represents a predetermined attribute value of that edge. Possible edge attribute values ​​include the type of chemical bond (single bond, double bond, etc.), the presence or absence of rings, and the presence or absence of conjugation. For example, in the case of the monomer in Figure 3, there are three connecting edges, so its edge characteristic matrix has three rows, and the edge attribute values ​​to be contained are assigned to each column. An example of an edge characteristic matrix with three arbitrary edge attribute values ​​is shown below.

[0050]

number

[0051] (4) Characteristic matrix of polymer graph according to the present invention

[0052] 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 described above. For example, the polymer node feature matrix PF (Polymer Feature matrix), which is the feature matrix of the exemplary polymer graph in Figure 3, may also have additional attribute information in the third column, as shown in the example below, and this can be represented by an attribute value indicating whether or not each node is a connected node. For example, the polymer node feature matrix PNF below may have connected node attribute information in the fourth column, as shown below, in addition to the monomer node feature matrix NF which contains the three types of attribute information of the four nodes mentioned above. In the example in Figure 3, nodes 1 and 4 are connected nodes that are connected to adjacent repeating units, so they have a value of "1", and the other nodes are represented as having a value of "0" in the fourth column.

[0053]

number

[0054] This is accomplished using a polymer node variable setting step, in which each atom of the monomers forming the selected polymer and the connection nodes, which are the locations where the monomers are repeatedly connected, are set as polymer node variables; and a polymer graph node information assignment step, in which connection nodes connected to the set polymer node variables are assigned and node attribute values ​​are assigned 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, and in particular, in one embodiment of the present invention, the node may have a connection node attribute value which is an attribute value indicating the presence or absence of a node connected to an adjacent repeating unit value.

[0055] In another embodiment of the present invention, connection node information can be represented by an edge characteristic matrix. For example, instead of representing whether or not a node is connected by a node attribute value, as in the polymer node characteristic matrix described above, or in conjunction with it, the edge characteristic matrix can represent attribute values ​​indicating the presence or absence of a connected edge. In the following example representing Figure 3(b), it can be seen that the fourth and fifth rows of data have been added to accommodate the information of the edges connecting connection nodes 1 and 4, i.e., edges 5_1 and 4_6 in Figure 3(b), and an attribute information column indicating the presence or absence of a connected edge has been added to the fourth column.

[0056]

number

[0057] In other words, according to the present invention, the method for representing the structure of a polymer graphically and converting it into data usable by a computer can be selected from one or more of the following methods.

[0058] 1) In an adjacency matrix representing the connection relationships of each node in a repeating unit consisting of at least one monomer, the components of "connecting nodes" that are connected to adjacent repeating units are assigned as the "connection" attribute.

[0059] 2) A node attribute matrix representing the attributes of each node in the repeating unit is given a "connection status" attribute field, and the "connection status" attribute field of the "connected node" is assigned "connected".

[0060] 3) Add the edges connected to the connected node to the edge attribute matrix of the repeating unit to the connected edge, assign the edge attribute a "connected" attribute field, and assign "connected" to the field of the connected edge.

[0061] 2.2. Method for converting the chemical structure of polymers into data that can be processed by computers.

[0062] In this invention, when training an artificial neural network model described later and predicting the properties of a polymer substance using the trained artificial neural network, the chemical structure of the polymer substance must be converted into data that can be processed by a computer. This data conversion is carried out according to the method of this invention described below.

[0063] (1) Repeated unit information acquisition step

[0064] This step involves obtaining repeating unit information, which is the unit formed by the repetition of monomers to create a polymer. One of the features of the present invention is obtaining repeating unit information that forms a polymer in order to digitize the connection node information in the repeating unit. The repeating unit information involves confirming whether the repeating unit that forms a polymer is repeated as a single monomer or as a linkage of single monomers, and describing the repeating unit using a graphical representation. The repeating unit can be a single monomer as shown in Figure 2(a), or it can be a case where monomers are linked together as shown in Figure 2(b).

[0065] (2) Polymer graph node connection information assignment step

[0066] The present invention provides a method for processing polymer graph information, which involves setting as variables the connection relationships between nodes representing each atom constituting a repeating unit of a polymer selected as the target of analysis, and connection nodes connected to connection points, which are locations where the repeating units are repeatedly connected.

[0067] In this way, attribute values ​​representing the interconnections between the node and the connected nodes are assigned to the polymer node connection variables. The polymer adjacency matrix, which was given as an example above, can be used to assign attribute values ​​for each interconnection to the polymer node connection variables.

[0068] (3) Polymer graph node attribute information assignment step

[0069] Furthermore, the polymer graph information processing method of the present invention may include a polymer node attribute variable setting step in which attribute information of nodes corresponding to each atom constituting the repeating unit that forms the selected polymer, and attribute information of connection nodes connected to connection points, which are locations where the repeating unit is repeatedly connected, are set as polymer node attribute variables.

[0070] The polymer graph node attribute information assignment step further includes assigning attribute values ​​of the node and connected nodes to the polymer node attribute variables set in this manner. The node attribute information has an attribute value indicating whether the node is a connected node or not, and as node attribute information for a connected node, a connected node attribute value is given that indicates that, unlike a node that is not a connected node, it is connected to a connected node of an adjacent repeating unit within the polymer.

[0071] (4) Polymer graph edge attribute information assignment step

[0072] Furthermore, the polymer graph information processing method of the present invention may include a polymer edge attribute variable setting step, in which information of the edges that connect nodes corresponding to each atom constituting the repeating unit forming the selected polymer, and the connecting edges that connect at least one of the nodes to the connecting node, is set as polymer edge attribute variables. A connecting edge is an edge that connects a connecting node to another node.

[0073] Furthermore, the present invention includes a polymer graph edge attribute information assignment step in which attribute values ​​of the edge and connected edge are assigned to the set polymer edge attribute variable, wherein a connected edge is assigned a connected edge attribute value as its edge attribute variable, which means that the connected node connects to other nodes.

[0074] 3. Method for predicting the glass transition temperature of a polymer using a graphical representation of the polymer according to the present invention.

[0075] The following describes a method for processing polymer data in an artificial intelligence neural network according to the polymer graph representation method of the present invention described above.

[0076] The present invention provides a method for predicting the glass transition temperature of a polymer by constructing an artificial neural network model that predicts the glass transition temperature of a polymer by applying the above-described graph representation method of polymers and using machine learning.

[0077] 3.1. Method for generating a predictive model for the glass transition temperature of polymers

[0078] Figure 4(a) shows a method for generating a polymer glass transition temperature prediction model according to the present invention. Based on this, a method for generating a computer-implemented polymer glass transition temperature prediction model that predicts the glass transition temperature of a selected polymer from its chemical structure, according to the graph information representation method of the present invention, will be described.

[0079] The polymer glass transition temperature prediction model of the present invention is configured to predict the glass transition temperature by applying the relationship between the polymer glass transition temperature and molecular weight, as defined by the Florey-Fox equation in Equation 1. g The glass transition temperature is a physical property that is influenced by both molecular structure and molecular weight, but conventional techniques have been unable to reflect both the influence of molecular structure and molecular weight when predicting the glass transition temperature. This invention realizes a prediction model with even greater accuracy by reflecting both molecular structure and molecular weight in the glass transition temperature prediction using the following formula 1. Note that for polymers with the same molecular structure, the T due to changes in molecular weight is... g It has become possible to predict this.

[0080]

number

[0081] (T g,infK is the glass transition temperature when the molecular weight is theoretically infinite, K is an experimental parameter related to the free volume within the polymer, and M n (This is the number-average molecular weight of the polymer.)

[0082] Therefore, the predictive model of the present invention uses the molecular structure of the selected polymer to determine T in Equation 1 above. g,inf It includes an artificial neural network that predicts T and K, and the output of the artificial neural network is T g,inf Applying the K value and the molecular weight of the selected polymer to Equation 1 above, the glass transition temperature T of the selected polymer is obtained. g It is configured to calculate the glass transition temperature characteristic information of the polymer, T g,inf The following procedure is performed to generate an artificial neural network model that predicts the K value.

[0083] (1) Preparation of training data (T10)

[0084] To generate the above-mentioned polymer glass transition temperature prediction model, training data is prepared for training an artificial neural network. The training data of this invention includes polymer graph information including the structural information of the sample polymer, and measurement data of the sample polymer. g Value, molecular weight M of the sample polymer n It consists of the dataset}.

[0085] Furthermore, the present invention includes, as training data, polymer graph information including the structural information of the sample polymer, and measurement data of the sample polymer. g Value, molecular weight M of the sample polymer n The dataset includes a set of}, but is characterized by including a predetermined number or more datasets of polymers having similar structures but different molecular weights, and datasets of polymers having different structures but the same molecular weight. By configuring the training dataset in this way, the artificial neural network of the present invention, described later, predicts T g Value (T gThis allows the predicted values ​​to be learned to reflect both the structural and molecular weight characteristics of the polymer.

[0086] (2) Construction of an artificial neural network and input of training data (T20)

[0087] Next, an artificial neural network model to be trained is constructed using the training data. In this invention, known artificial neural networks can be used, and in particular, message-passing-based MPNNs (message passing neural networks) and GNNs (graph neural networks) can be applied.

[0088] For example, when applying a GNN as an artificial neural network, in the case of supervised learning, the graph neural network (GNN) includes the polymer graph information, including the structural information of the sample polymer, and the measurement of the sample polymer T. g Value, molecular weight M of the sample polymer n Input molecular training data consisting of pairs of}.

[0089] At this time, the polymer graph information is input as the input value to the artificial neural network, and the known measurement T of the polymer is performed. g Value and molecular weight M n This is input as the true value (labeled data) of the neural network's output. In this case, the input polymer graph information includes at least one of the polymer graph node connection information, polymer graph node attribute information, and polymer graph edge attribute information mentioned above, and can be input embedded in the polymer adjacency matrix and polymer node / edge characteristic matrix.

[0090] (3) Learning steps of an artificial neural network (T30)

[0091] This is the process of updating the neural network parameters of the artificial neural network and learning the artificial neural network based on the input polymer graph information, molecular weight information, and polymer glass transition temperature characteristic information. The learning process of the artificial neural network conforms to known artificial neural network learning processes. For example, the measured molecular weight information of the learning dataset is reflected in the predicted value of the polymer glass transition temperature characteristic information calculated from the artificial neural network upon input of the polymer graph information of the learning data, according to the above formula 1. g Calculate the predicted value, T g Predicted value and measurement T of the polymer in question from the training data g The neural network parameters of the artificial neural network are updated to minimize the error function, which is defined as the difference between the value and the target value. By updating the neural network parameters in this way, an artificial neural network capable of calculating glass transition temperature characteristic information from the structure of a polymer is completed.

[0092] Thus, the present invention reflects molecular weight information. g The artificial neural network is trained to minimize the error between the predicted value and the measured value, thus enabling accurate T in polymers. g It is possible to make predictions.

[0093] On the other hand, in this invention, the term "neural network parameters" of an artificial neural network is used in the common sense of the art to refer to the weights and biases that are adjusted as the artificial neural network model learns from training data, and that such parameters are adjusted so that the neural network model predicts a more accurate output from the input data, but are optimized to predict the best possible output during the machine learning process.

[0094] 3.2. Method for predicting polymer glass transition temperature

[0095] The graph information of the selected polymer is input into the artificial neural network generated according to the polymer glass transition temperature model generation method of the present invention, and the number average molecular weight M of the selected polymer is input. n Reflecting this, we will describe a procedure for predicting the glass transition temperature of a selected polymer.

[0096] (1) Preprocessing step for polymer graph information (S10)

[0097] As shown in Figure 4(b), a preprocessing step is performed to obtain polymer graph information from the chemical structure of the selected polymer whose glass transition temperature is to be predicted. As mentioned above, the polymer graph information includes node and edge information of the monomers constituting the polymer, as well as information on connection nodes and connection edges 40 between repeating monomers. That is, the polymer graph information is preprocessed to include at least one of the polymer graph node connection information, polymer graph node attribute information, and polymer graph edge attribute information described above.

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

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

[0100] (3) Prediction step of polymer glass transition temperature characteristics (S30)

[0101] From the polymer graph information of the selected polymer, the glass transition temperature characteristic information of the selected polymer (T g,inf This step involves calculating and outputting the K value using the trained artificial neural network model.

[0102] (4) Step to reflect molecular weight information (S40)

[0103] Glass transition temperature characteristic information (T g,inf Using the above formula 1 (and K value), the glass transition temperature T g This is the step to calculate the molecular weight of the polymer. The molecular weight of the polymer is obtained and reflected by a known method.

[0104] 4. A system for calculating the glass transition temperature of polymers using a graphical representation of polymers according to the present invention.

[0105] (1) Data input section 100

[0106] The data input unit 100 is a component that receives polymer graph information, which describes the chemical structure of a selected polymer using graphical data, and molecular weight data of the polymer. It further includes a data preprocessing unit (not shown) that converts the chemical structure of the polymer into polymer graph information and inputs this into the data input unit. Molecular weight data (molecular weight information) can be obtained by known methods.

[0107] When inputting polymer graph information into the artificial neural network model unit described later, the data preprocessing unit may generate and input a polymer adjacency matrix that represents the connection relationships between two or more nodes representing each atom constituting the single molecule and the connection relationship information between the connected nodes, one or more edges representing the connections between each node, and attribute information of the connection edges representing the connections between the nodes.

[0108] In this case, the polymer graph information may include, as described above, polymer graph node connection information which represents the connection relationships between nodes corresponding to each atom constituting the repeating unit that forms the polymer and connection nodes that are connected to connection points, which are locations where the repeating unit is repeatedly connected; polymer node attribute information which represents the attributes of the nodes corresponding to each atom constituting the repeating unit that forms the polymer and connection nodes that are connected to connection points, which are locations where the repeating unit is repeatedly connected; and polymer edge attribute information which represents the attributes of the edges that connect the nodes corresponding to each atom constituting the repeating unit that forms the polymer and connection edges that connect at least one of the nodes to the connection node.

[0109] (2) Artificial neural network model unit 200

[0110] This component predicts and calculates the glass transition temperature characteristic information of the polymer from the polymer graph information. The glass transition temperature characteristic information of the polymer is T in Equation 1 above. g,inf This corresponds to K.

[0111] Such an artificial neural network model unit is configured to receive polymer graph information as input and output a predicted value of the glass transition temperature characteristic information of the polymer. It calculates a predicted value of the glass transition temperature characteristic information of a predetermined learning polymer from the polymer graph information of the learning polymer, and predicts T from the calculated predicted value of the glass transition temperature characteristic information of the learning polymer and the measured molecular weight information of the learning polymer. g Calculate the value and predict T g Value and measurement T of the polymer in question among the training data g The system includes an artificial neural network for predicting polymer glass transition temperature characteristics, which is trained by updating its neural network parameters to minimize the error function, defined as the difference between the value and the actual value.

[0112] (3) Glass transition temperature calculation unit 300

[0113] The glass transition temperature calculation unit is a component that calculates the glass transition temperature of a polymer according to the above-described formula 1, using the polymer glass transition temperature characteristic information calculated by the artificial neural network model unit and the polymer molecular weight data received from the data input unit.

[0114] The above describes a method for graphically representing the structure of a polymer using graph information, a method and system for predicting the glass transition temperature characteristics of a polymer from the graph information of the polymer, and finally calculating the accurate glass transition temperature of the polymer by reflecting the molecular weight information of the polymer in the predicted glass transition temperature characteristics.

[0115] According to the method and system for calculating the glass transition temperature of a polymer that reflects molecular weight information according to the present invention, it becomes possible to consider the effect of molecular weight separately and adjust the prediction according to the molecular weight, thereby enabling the construction of an accurately predictable model.

[0116] For example, actual T g When samples are similar, conventional models only reflect the molecular structure without considering molecular weight, and therefore predict the molecular weight of samples with similar structures regardless of molecular weight. g The system learns and predicts values ​​to be similar.

[0117] TIFF0007881747000010.tif79170

[0118] TIFF0007881747000011.tif51170

[0119] TIFF0007881747000012.tif79170

[0120] TIFF0007881747000013.tif41170

[0121] Thus, according to the present invention, the neural network model is trained to predict the glass transition temperature by reflecting not only the molecular structure of the polymer but also its molecular weight information. As a result, the influence of both molecular structure information and molecular weight information is learned, making it possible to predict the glass transition temperature with even greater accuracy compared to conventional techniques. [Explanation of Symbols]

[0122] 10 nodes 20 Edge 30 connection points 40 Connected Edges 100 Data Input Section 200 Artificial Neural Network Model Section 300 Glass transition temperature calculation unit

Claims

1. A method for generating a computer-implemented model that predicts the glass transition temperature of a selected polymer, A training data acquisition step involves preparing a training dataset containing structural information of multiple polymers for training, known glass transition temperature measurements of the polymers, and measured molecular weights of the polymers. A polymer glass transition temperature characteristic information prediction artificial neural network is constructed to receive polymer structure information as input and output predicted values ​​of Tg, inf, and K values ​​in the following formula 1 as glass transition temperature characteristic information of the polymer, in a polymer glass transition temperature characteristic network construction step, A learning data input step involves inputting the structural information of the polymer for learning data and the measured glass transition temperature of the polymer for learning data into the polymer glass transition temperature characteristic information prediction artificial neural network, The process includes a learning step for a polymer glass transition temperature prediction artificial neural network, in which the neural network parameters of the polymer glass transition temperature prediction artificial neural network are updated based on a comparison result between the predicted glass transition temperature of the polymer for training data, which is output by the polymer glass transition temperature prediction artificial neural network, and the predicted glass transition temperature of the polymer for training data, which is calculated by applying the measured molecular weight of the polymer in the training dataset to the following formula 1, and the measured glass transition temperature of the polymer for training data. The polymer structure information is converted into polymer graph information and input into the polymer glass transition temperature characteristic prediction artificial neural network. The polymer graph information includes polymer graph node connection information, which represents the connection relationships between nodes corresponding to each atom constituting the repeating unit that forms the polymer and connection nodes connected to connection points, which are locations where the repeating unit is repeatedly connected; attribute information of the nodes corresponding to each atom constituting the repeating unit; polymer node attribute information, which represents the attributes of the connection nodes; attributes of the edges connecting the nodes corresponding to each atom constituting the repeating unit to each other; and polymer edge attribute information, which represents the attributes of the connection edges connecting at least one of the nodes to the connection node. A method for generating a predictive model for the glass transition temperature of polymers. [Number 10] (Tg is the glass transition temperature of the polymer with molecular weight Mn, Tg,inf is the glass transition temperature when the molecular weight is infinite, K is the free volume relation of the polymer, and Mn is the number-average molecular weight of the polymer.)

2. A computer implementation method for calculating a predicted glass transition temperature of a selected polymer, A step to acquire a polymer glass transition temperature characteristic information prediction artificial neural network is provided, which is machine-trained to predict the Tg, inf, and K values ​​in the following formula 1 as glass transition temperature characteristic information of the polymer, based at least partially on the chemical structure related to the selected polymer. A polymer graph information acquisition step, which involves acquiring polymer graph information that describes the chemical structure of the selected polymer using graphic data, A polymer graph input step provides the acquired polymer graph information as input to the graph neural network, A polymer glass transition temperature characteristic information acquisition step, which involves receiving polymer glass transition temperature characteristic information of the selected polymer from the output of the graph artificial neural network, The step includes calculating the glass transition temperature of the selected polymer by substituting the molecular weight information of the selected polymer into the following formula 1 for the polymer glass transition temperature characteristic information, and then performing the calculation of the glass transition temperature of the selected polymer. The polymer graph information includes polymer graph node connection information, which represents the connection relationships between nodes corresponding to each atom constituting the repeating unit that forms the polymer and connection nodes connected to connection points, which are locations where the repeating unit is repeatedly connected; attribute information of the nodes corresponding to each atom constituting the repeating unit; polymer node attribute information, which represents the attributes of the connection nodes; attributes of the edges connecting the nodes corresponding to each atom constituting the repeating unit to each other; and polymer edge attribute information, which represents the attributes of the connection edges connecting at least one of the nodes to the connection node. Methods for realizing a computer. [Math 11] (Tg is the glass transition temperature of the polymer with molecular weight Mn, Tg,inf is the glass transition temperature when the molecular weight is infinite, K is the free volume relation variable of the polymer, and Mn is the number-average molecular weight of the polymer.)

3. The polymer graph input step, which provides the acquired polymer graph information as input to the machine-learned graph neural network, A polymer adjacency matrix construction step, which involves constructing an adjacency matrix that represents the mutual connection relationships between two or more nodes representing each atom constituting the repeating unit, The step includes inputting a polymer adjacency matrix, in which the constructed graph matrix is ​​input as the graph neural network, The aforementioned polymer adjacency matrix is, The computer implementation method according to claim 2, further comprising connection relationship information between the connection node and the node.

4. The aforementioned graph artificial neural network is, From the structural information of multiple training polymers, we calculate the predicted glass transition temperature characteristic information of the training polymers. The computer implementation method according to claim 2, which is learned by updating the neural network parameters to minimize the error between the predicted glass transition temperature, calculated by reflecting the measured molecular weight of the polymer in the calculated predicted glass transition temperature characteristic information, and the measured glass transition temperature of the polymer used for training data.

5. The step of acquiring the polymer glass transition temperature characteristic information prediction artificial neural network is as follows: The steps include: preparing training data using one or more computing devices, which includes the chemical structures of multiple exemplary polymers, as well as measured glass transition temperatures and molecular weights of the exemplary polymers having the chemical structures of the exemplary polymers; The computer implementation method according to claim 2, comprising a learning step of acquiring graph information that graphically describes the chemical structure of the exemplary polymer, inputting it into a graph neural network, and training the graph neural network so that it outputs a predicted value of the glass transition temperature characteristic information of the exemplary polymer.

6. The graphical information describing the chemical structure of the aforementioned exemplary polymer is: Two or more node information representing each atom constituting the repeating unit that makes up the polymer, One or more edge pieces representing the bonds between each of the aforementioned atoms, The connection node information is the location where the repeating unit is repeatedly connected, The computer implementation method according to claim 5, comprising connection edge information representing the connection between the atom and the connection node.

7. A data input unit that receives the structural information of the selected polymer and the molecular weight data of the polymer, An artificial neural network model unit calculates predicted values ​​of Tg, inf, and K values ​​in the following formula 1 as glass transition temperature characteristic information of the polymer from the structural information of the polymer. The system comprises a glass transition temperature calculation unit that calculates a predicted glass transition temperature of the polymer from the predicted glass transition temperature characteristic information of the polymer and the molecular weight data of the polymer. The aforementioned structural information is converted into polymer graph information and input into the artificial neural network model unit. The polymer graph information includes polymer graph node connection information, which is information representing the connection relationships between nodes corresponding to each atom constituting the repeating unit that forms the polymer and connection nodes connected to connection points, which are locations where the repeating unit is repeatedly connected among the nodes; attribute information of the nodes corresponding to each atom constituting the repeating unit; polymer node attribute information, which is information representing the attributes of the connection nodes; attributes of the edges connecting the nodes corresponding to each atom constituting the repeating unit to each other; and polymer edge attribute information, which is information representing the attributes of the connection edges connecting at least one of the nodes to the connection node. The glass transition temperature calculation unit calculates the glass transition temperature Tg from the following formula 1. A system for calculating the glass transition temperature of polymers. [Math 12] (Tg is the glass transition temperature of the polymer with molecular weight Mn, Tg,inf is the glass transition temperature when the molecular weight is infinite, K is the free volume relation variable of the polymer, and Mn is the number-average molecular weight of the polymer.)

8. The system further includes a data preprocessing unit that converts the structural information of the polymer into polymer graph information. The polymer glass transition temperature calculation system according to claim 7, wherein the artificial neural network model unit is machine-trained to calculate a predicted value of the glass transition temperature characteristic information of the polymer from the polymer graph information.

9. The artificial neural network model unit is, The system comprises an artificial neural network for predicting polymer glass transition temperature characteristics, which is constructed to receive polymer structure information as input and output predicted values ​​of Tg, inf, and K values ​​in the above formula 1 as glass transition temperature characteristic information of the polymer. The aforementioned artificial neural network for predicting polymer glass transition temperature characteristics is: Using the structural information of a predetermined polymer for training data, and the measured glass transition temperature and molecular weight of the polymer for training data as training data, From the structural information of the polymer used for training data, a predicted value of the glass transition temperature characteristic information of the polymer used for training data is calculated. A polymer glass transition temperature calculation system according to claim 7, wherein the predicted glass transition temperature characteristic information and the measured molecular weight are applied to the formula 1 to calculate the predicted glass transition temperature of the polymer for training data, and the parameters of the artificial neural network are updated by comparing the measured glass transition temperature of the polymer for training data with the calculated value to perform machine learning.

10. The aforementioned training data is Polymer graph information including structural information of the sample polymer, and measurement T of the sample polymer. g Value and molecular weight M of the sample polymer. n Includes a dataset containing, At least a predetermined number of the aforementioned datasets are A polymer glass transition temperature calculation system according to claim 9, comprising a dataset of polymers having similar structures but different molecular weights, and a dataset of polymers having different structures but the same molecular weight.

Images