Molecular property prediction method and device based on graph convolutional neural network

Abstract

This invention provides a method and apparatus for predicting molecular properties based on graph convolutional neural networks. The method includes: constructing a target molecular graph with atoms as nodes and a target substructure graph with molecular substructures as nodes based on the molecular graph identifier of the target molecule; extracting target molecular graph features from the target molecular graph and target molecular substructure features from the target substructure graph; fusing the target molecular graph features and target molecular substructure features to obtain the overall features of the target molecule; and inputting the overall features of the target molecule into a graph convolutional neural network model to obtain the molecular properties of the target molecule output by the graph convolutional neural network model. The method and apparatus for predicting molecular properties based on graph convolutional neural networks provided by this invention extract structural features at different levels of the target molecule to obtain structural information at different levels, and then accurately predict the molecular properties of the target molecule based on this structural information at different levels.

Description

Technical Field

[0001] This invention relates to the fields of petrochemical technology and artificial intelligence algorithm technology, and in particular to a method and apparatus for predicting molecular properties based on graph convolutional neural networks. Background Technology

[0002] With the continuous development of artificial intelligence and big data in recent years, the petrochemical industry, as one of the pillar industries of energy, is also attempting to integrate with AI-related technologies and develop towards intelligentization. Molecular management is one of the important means to optimize crude oil resources and improve crude oil utilization. Among them, establishing the structure-property relationship between molecular structure and molecular properties, that is, predicting molecular properties through molecular structure, is a fundamental and important link in the field of molecular management.

[0003] In the process of constructing a crude oil molecular model, the group contribution method is needed to calculate the relevant basic physical properties of a predefined molecule. First, its structure is identified and disassembled, and then a mathematical model is established to calculate the properties. The group contribution method is a relatively traditional method for predicting properties through molecular structure. It requires first extracting typical structural groups that affect the properties, and then predicting molecular properties based on the number and connection methods of these groups. This process is heavily reliant on structural division weights, and the method has poor scalability. Summary of the Invention

[0004] This invention provides a method and apparatus for predicting molecular properties based on graph convolutional neural networks, in order to overcome the deficiencies in the prior art.

[0005] This invention provides a method for predicting molecular properties based on graph convolutional neural networks, comprising the following steps: Based on the molecular graph identifier of the target molecule, construct the target molecular graph with atoms as nodes and the target substructure graph with molecular substructure as nodes; Extract target molecular map features from the target molecular map, and extract target molecular substructure features from the target substructure map; By fusing the target molecule map features and the target molecule substructure features, the overall features of the target molecule are obtained; The overall features of the target molecule are input into a graph convolutional neural network model to obtain the molecular properties of the target molecule output by the graph convolutional neural network model.

[0006] According to the present invention, a method for predicting molecular properties based on graph convolutional neural networks is provided, wherein constructing a target molecular graph with atoms as nodes and a target substructure graph with molecular substructures as nodes based on the molecular graph identifier of the target molecule includes: Based on the molecular graph identifier, the target molecular graph is constructed with atoms as nodes and chemical bonds as edges; Based on the molecular graph identifier, the target substructure graph is constructed with molecular substructures as nodes and atoms as edges.

[0007] According to the present invention, a method for predicting molecular properties based on graph convolutional neural networks includes extracting target molecular graph features from the target molecular graph and extracting target molecular substructure features from the target substructure graph, comprising: Based on the feature extraction model, target molecular graph features are extracted from the target molecular graph, and target molecular substructure features are extracted from the target substructure graph.

[0008] According to the present invention, a molecular property prediction method based on graph convolutional neural networks is provided, wherein the molecular graph identifier is determined based on the simplified molecular input line representation identifier of the target molecule.

[0009] According to the present invention, a method for predicting molecular properties based on a graph convolutional neural network is provided, wherein the graph convolutional neural network model is trained based on the following steps: Obtain the target datasets corresponding to different training tasks; Each target dataset is divided into a training set and a test set; Based on the training set of each target dataset, the parameters of the initial model of the graph convolutional neural network model are updated to obtain the graph convolutional neural network model.

[0010] According to the present invention, a molecular property prediction method based on graph convolutional neural network is provided. Each target dataset includes the chemical formula, structural formula and important physicochemical properties of sample molecules. The important physicochemical properties include at least one of boiling point, critical temperature, critical pressure, molar volume, melting point, Gibbs free energy, eccentricity factor, standard enthalpy of formation and solubility parameter.

[0011] The present invention also provides a molecular property prediction device based on graph convolutional neural networks, comprising the following modules: Construction units are used to construct target molecular graphs with atoms as nodes and target substructure graphs with molecular substructures as nodes, based on the molecular graph identifier of the target molecule. An extraction unit is used to extract target molecular map features from the target molecular map and to extract target molecular substructure features from the target substructure map; A fusion unit is used to fuse the target molecule map features and the target molecule substructure features to obtain the overall features of the target molecule. The prediction unit is used to input the overall features of the target molecule into the graph convolutional neural network model to obtain the molecular properties of the target molecule output by the graph convolutional neural network model.

[0012] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the molecular property prediction method based on graph convolutional neural networks as described above.

[0013] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the molecular property prediction method based on graph convolutional neural networks as described above.

[0014] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the molecular property prediction method based on graph convolutional neural networks as described above.

[0015] The molecular property prediction method and apparatus based on graph convolutional neural networks provided by this invention extracts structural features (graph features and substructure features) of the target molecule at different levels to obtain structural information of the target molecule at different levels, and then accurately predicts the molecular properties of the target molecule based on the structural information at different levels. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating the molecular property prediction method based on graph convolutional neural networks provided by this invention.

[0018] Figure 2 This is a schematic diagram of the training process of the graph convolutional neural network model provided by the present invention.

[0019] Figure 3 This is a schematic diagram of vector extraction provided by the present invention.

[0020] Figure 4 This is a schematic diagram of the molecular property prediction device based on graph convolutional neural networks provided by the present invention.

[0021] Figure 5 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0023] In the process of constructing a crude oil molecular model, the group contribution method is needed to calculate the relevant basic physical properties of a predefined molecule. First, its structure is identified and disassembled, and then a mathematical model is established to calculate the properties. The group contribution method is a relatively traditional method for predicting properties through molecular structure. It requires first extracting typical structural groups that affect the properties, and then predicting molecular properties based on the number and connection methods of these groups. This process is heavily reliant on structural division weights, and the method has poor scalability.

[0024] In recent years, with the rapid development of deep learning technology and the continuous accumulation of related data, it has become possible to predict molecular properties through data mining. In undirected molecular graphs, atoms serve as graph nodes, and chemical bonds as edges connecting the nodes. Using graph representation methods can preserve molecular structure and topological connection information to the greatest extent. However, molecular property prediction requires a very large sample size of the dataset. Therefore, this invention constructs a framework for small-sample molecular property prediction based on a graph convolutional neural network model, improving the accuracy of molecular property prediction. This framework can be used to predict many different types of molecular properties, greatly enhancing the scalability and accuracy of molecular property prediction methods.

[0025] Figure 1 This is a flowchart illustrating the molecular property prediction method based on graph convolutional neural networks provided by this invention, as shown below. Figure 1 As shown, the method includes steps 110, 120, 130 and 140.

[0026] Step 110: Based on the molecular graph identifier of the target molecule, construct the target molecular graph with atoms as nodes and the target substructure graph with molecular substructures as nodes.

[0027] Specifically, the target molecule is the molecule whose properties are to be predicted. The molecular graph identifier of the target molecule can be determined based on the simplified molecular input line representation identifier (i.e., the SMILES expression identifier). The target molecular graph is constructed based on the molecular graph identifier, with atoms as nodes. In this graph, nodes represent atoms, and edges represent chemical bonds between atoms. The target substructure graph is constructed based on the molecular graph identifier, with molecular substructures as nodes. This graph captures higher-level structural information in the molecule; nodes represent molecular substructures (such as functional groups), and edges represent common atoms in adjacent molecular substructures.

[0028] Step 120: Extract target molecular map features from the target molecular map and extract target molecular substructure features from the target substructure map.

[0029] Specifically, target molecular map features are extracted from the target molecular map. These features may include atom types, chemical bond types, atom charges, bond lengths, bond angles, etc. These features are used to describe the local structure and chemical environment of the molecule.

[0030] Extracting target molecule substructure features from the target substructure map, these features may include the type, number, and location of substructures. These substructure features provide higher-level structural information about the molecule.

[0031] In other words, target molecule map features and target substructure features are features extracted from the structure of the target molecule at different levels, and thus the structural information of the target molecule at different levels can be obtained based on the target molecule map features and target molecule substructure features.

[0032] Step 130: Integrate the target molecule map features and the target molecule substructure features to obtain the overall features of the target molecule.

[0033] Specifically, the target molecule's graph features and substructure features are fused to obtain the overall features of the target molecule. This can be achieved through strategies such as feature splicing, weighted summation, and attention mechanisms to fully extract information from both the target molecule graph and its substructure features. Since both the target molecule graph features and substructure features can characterize different levels of structural information within the target molecule, the overall features of the target molecule can also characterize these different levels of structural information.

[0034] Step 140: Input the overall features of the target molecule into the graph convolutional neural network model to obtain the molecular properties of the target molecule output by the graph convolutional neural network model.

[0035] Specifically, the overall features of the fused target molecule are input into a graph convolutional neural network (GCN) model. The GCN model predicts the molecular properties of the target molecule based on the structural information at different levels carried in the overall features of the target molecule. The molecular properties can include physical properties (such as melting point and boiling point) and chemical properties (such as Gibbs free energy and eccentricity factor).

[0036] The molecular property prediction method based on graph convolutional neural networks provided in this invention extracts structural features (graph features and substructure features) at different levels of the target molecule to obtain structural information at different levels of the target molecule, and then accurately predicts the molecular properties of the target molecule based on the structural information at different levels.

[0037] Based on the above embodiments, based on the molecular graph identifier of the target molecule, a target molecular graph with atoms as nodes and a target substructure graph with molecular substructures as nodes are constructed, including: Based on molecular graph identification, the target molecular graph is constructed with atoms as nodes and chemical bonds as edges; Based on molecular graph identification, a target substructure graph is constructed using molecular substructures as nodes and atoms as edges.

[0038] Specifically, based on the molecular graph identifier, a target molecular graph is constructed with atoms as nodes to learn the structure-property relationships at the atomic level. In the target molecular graph, atoms are represented by nodes, and chemical bonds are represented by edges. For example, the target molecular graph can be described as M = (V, E), where V represents all the nodes represented by atoms in the target molecule, E represents the edges connecting the atomic nodes in the target molecule, and the features of the atomic nodes can be represented by one-hot encoded binary vectors.

[0039] Based on the molecular graph identifier, a target substructure graph is constructed with molecular substructures as nodes, and the structure-property relationships at the substructure level are learned. The target molecular substructure features can be extracted from the target substructure graph using a feature extraction model, where molecular substructures are treated as nodes and common atoms as edges. For example, the target substructure graph can be described as T = (S, A), where S corresponds to all substructures within the target molecule, and A corresponds to the common atoms of adjacent substructures.

[0040] Based on any of the above embodiments, extracting target molecular map features from the target molecular map and extracting target molecular substructure features from the target substructure map includes: Based on the feature extraction model, target molecular graph features are extracted from the target molecular graph, and target molecular substructure features are extracted from the target substructure graph.

[0041] Specifically, the target molecular map can be input into the feature extraction model, which will then extract the target molecular map features. Similarly, the target molecular structure map can be input into the feature extraction model, which will then extract the target molecular substructure features.

[0042] After obtaining the target molecular map features and target molecular substructure features, the extracted features can be standardized or normalized, and redundant features can be removed.

[0043] Based on any of the above embodiments, the molecular graph identifier is determined based on the simplified molecular input line representation identifier of the target molecule (i.e., the SMILES expression identifier).

[0044] Based on any of the above embodiments, the graph convolutional neural network model is trained using the following steps: Obtain the target datasets corresponding to different training tasks; Each target dataset is divided into a training set and a test set; Based on the training set of each target dataset, the parameters of the initial model of the graph convolutional neural network model are updated to obtain the graph convolutional neural network model.

[0045] Here, different training tasks can be understood as prediction tasks of different natures. For example, different training tasks may include predicting the physical properties of molecule a, predicting the chemical properties of molecule b, and predicting the chemical properties of molecule a, etc.

[0046] The target dataset can include the chemical formulas, structural formulas, and important physicochemical properties of different sample molecules. For example, the molecular database in Aspen Hysys can be used to collect typical molecules contained in crude oil, including their chemical formulas, structural formulas, and important physicochemical properties such as boiling point, critical temperature, critical pressure, molar volume, melting point, Gibbs free energy, eccentricity factor, standard enthalpy of formation, and solubility parameters, to obtain the corresponding target dataset.

[0047] Next, the target dataset is divided into a training set and a test set. The training set is used to train the model, and the test set is used to test the performance of the trained model. The training set and the test set are further divided into a support set and a query set, respectively. For example, random sampling can be performed on the training set to obtain a support set containing C×k samples for C training tasks (i.e., the molecular properties to be predicted). Samples are then taken from the remaining samples to form the query set, and the sum of the support set and the query set constitutes the training set. The test set is processed similarly.

[0048] The initial model parameters θ are updated iteratively multiple times using training sets corresponding to different training tasks to obtain the graph convolutional neural network model.

[0049] Therefore, the embodiments of the present invention optimize model parameters by using training sets corresponding to multiple different training tasks, thereby making up for the problem of insufficient training data caused by a single task, so that the trained model has a high accuracy in prediction of different properties.

[0050] Figure 2 This is a schematic diagram of the training process of the graph convolutional neural network model provided by the present invention, as shown below. Figure 2As shown, chemical molecular data is acquired to construct a molecular dataset. This chemical molecular data includes the chemical formula, structural formula, and important physicochemical properties of molecules, including boiling point, critical temperature, critical pressure, molar volume, melting point, Gibbs free energy, eccentricity factor, standard enthalpy of formation, and solubility parameters. The above molecular information is searched through the ASPEN Hysys database, and SMILES expression identifiers are output by identifying the molecular information. Molecular diagram identifiers are then output using the SMILES expression identifiers.

[0051] Figure 3 This is a schematic diagram of vector extraction provided by the present invention, such as... Figure 3 As shown, a molecular graph with atoms as nodes and a substructure graph with molecular substructures as nodes are constructed to learn the structure-property relationships at the atomic level and the substructure level. For atomic node i, its feature representation is as follows: The characteristic representation of a molecular diagram is as follows: Where N represents the number of atomic nodes in the molecular graph, and L represents the dimension of the eigenvector of each atomic node. The topological connections of the molecule are represented by an adjacency matrix. This indicates that if there is a connection between atomic node i and atomic node j, then in the corresponding adjacency matrix... If there is no chemical bond connecting atomic node i and atomic node j, then in the corresponding adjacency matrix... Furthermore, molecular structure determines molecular function, and functional groups are an important component of molecular structure and a crucial basis for understanding molecular properties. Using molecular substructures (functional groups) as basic nodes, substructure diagram descriptors are constructed for compound molecules. Molecular substructures consist of multiple atoms and chemical bonds between them. If there are connections between substructures, the common atoms contained in two adjacent substructures are recorded as edges in the molecular substructure diagram. Where N represents the number of substructures (functional groups) in the molecular diagram, and L represents the dimension of the substructure eigenvectors. If substructures share common atomic connections, then in the corresponding adjacency matrix... If there are no common atoms connected between substructures, then in the corresponding adjacency matrix... .

[0052] like Figure 3 As shown, the method for fusing molecular diagram embedding vectors and substructure diagram embedding vectors is as follows: ,[,] refers to the concatenate operation. This refers to the molecular graph embedding vector. Finger structure graph embedding vector, This refers to the final molecular feature vector.

[0053] This invention embodiment constructs a graph convolutional neural network transfer learning model and uses experimental data for transfer learning training. Specific operations include: constructing a transfer learning model framework; training multiple target data tasks based on this framework; iteratively updating parameters using sample data from different target data training tasks; and outputting model parameters θ. These parameters θ will be used for transfer training on small-sample tasks. Figure 2 As shown, this model framework trains multiple tasks simultaneously and updates the parameters multiple times based on the number of samples from different tasks, ultimately obtaining a model parameter θ with good performance.

[0054] The loss function is the molecular property prediction loss function, which is calculated using the cross-entropy formula, as follows: Where k represents the number of data points in the dataset. , These represent the actual values ​​and the model's predicted values, respectively. The dataset is divided into a training set and a test set.

[0055] Based on any of the above embodiments, each target dataset includes the chemical formula, structural formula, and important physicochemical properties of the sample molecules. The important physicochemical properties include at least one of the following: boiling point, critical temperature, critical pressure, molar volume, melting point, Gibbs free energy, eccentricity factor, standard enthalpy of formation, and solubility parameter.

[0056] The molecular property prediction device based on graph convolutional neural networks provided by the present invention will be described below. The molecular property prediction device based on graph convolutional neural networks described below can be referred to in correspondence with the molecular property prediction method based on graph convolutional neural networks described above.

[0057] Based on any of the above embodiments Figure 4 This is a schematic diagram of the molecular property prediction device based on graph convolutional neural networks provided by the present invention, as shown below. Figure 4 As shown, the device includes: Construction unit 410 is used to construct a target molecular graph with atoms as nodes and a target substructure graph with molecular substructures as nodes based on the molecular graph identification of the target molecule. Extraction unit 420 is used to extract target molecular map features from the target molecular map and target molecular substructure features from the target substructure map; Fusion unit 430 is used to fuse target molecule map features and target molecule substructure features to obtain the overall features of the target molecule; The prediction unit 440 is used to input the overall features of the target molecule into the graph convolutional neural network model to obtain the molecular properties of the target molecule output by the graph convolutional neural network model.

[0058] The molecular property prediction device based on graph convolutional neural network provided in this invention extracts structural features (graph features and substructure features) of the target molecule at different levels to obtain structural information of the target molecule at different levels, and then accurately predicts the molecular properties of the target molecule based on the structural information at different levels.

[0059] Based on any of the above embodiments, constructing a target molecular graph with atoms as nodes and a target substructure graph with molecular substructures as nodes, based on the molecular graph identifier of the target molecule, includes: Based on molecular graph identification, the target molecular graph is constructed with atoms as nodes and chemical bonds as edges; Based on molecular graph identification, a target substructure graph is constructed using molecular substructures as nodes and atoms as edges.

[0060] Based on any of the above embodiments, extracting target molecular map features from the target molecular map and extracting target molecular substructure features from the target substructure map includes: Based on the feature extraction model, target molecular graph features are extracted from the target molecular graph, and target molecular substructure features are extracted from the target substructure graph.

[0061] Based on any of the above embodiments, the molecular graph identifier is determined based on the simplified molecular input line representation identifier of the target molecule.

[0062] Based on any of the above embodiments, the graph convolutional neural network model is trained using the following steps: Obtain the target datasets corresponding to different training tasks; Each target dataset is divided into a training set and a test set; Based on the training set of each target dataset, the parameters of the initial model of the graph convolutional neural network model are updated to obtain the graph convolutional neural network model.

[0063] Based on any of the above embodiments, each target dataset includes the chemical formula, structural formula, and important physicochemical properties of the sample molecules. The important physicochemical properties include at least one of the following: boiling point, critical temperature, critical pressure, molar volume, melting point, Gibbs free energy, eccentricity factor, standard enthalpy of formation, and solubility parameter.

[0064] Figure 5 This is a schematic diagram of the structure of the electronic device provided by the present invention, such as... Figure 5As shown, the electronic device may include a processor 510, a communications interface 520, a memory 530, and a communication bus 540, wherein the processor 510, communications interface 520, and memory 530 communicate with each other via the communication bus 540. The processor 510 can call logical instructions in the memory 530 to execute a molecular property prediction method based on a graph convolutional neural network. This method includes: constructing a target molecular graph with atoms as nodes and a target substructure graph with molecular substructures as nodes based on the molecular graph identifier of the target molecule; extracting target molecular graph features from the target molecular graph and extracting target molecular substructure features from the target substructure graph; fusing the target molecular graph features and the target molecular substructure features to obtain the overall features of the target molecule; and inputting the overall features of the target molecule into a graph convolutional neural network model to obtain the molecular properties of the target molecule output by the graph convolutional neural network model.

[0065] Furthermore, the logical instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0066] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute the molecular property prediction method based on graph convolutional neural networks provided by the above methods. The method includes: constructing a target molecular graph with atoms as nodes and a target substructure graph with molecular substructures as nodes based on the molecular graph identifier of the target molecule; extracting target molecular graph features from the target molecular graph and extracting target molecular substructure features from the target substructure graph; fusing the target molecular graph features and the target molecular substructure features to obtain the overall features of the target molecule; and inputting the overall features of the target molecule into a graph convolutional neural network model to obtain the molecular properties of the target molecule output by the graph convolutional neural network model.

[0067] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the molecular property prediction method based on graph convolutional neural networks provided by the above methods. This method includes: constructing a target molecular graph with atoms as nodes and a target substructure graph with molecular substructures as nodes based on the molecular graph identifier of the target molecule; extracting target molecular graph features from the target molecular graph and extracting target molecular substructure features from the target substructure graph; fusing the target molecular graph features and the target molecular substructure features to obtain the overall features of the target molecule; and inputting the overall features of the target molecule into a graph convolutional neural network model to obtain the molecular properties of the target molecule output by the graph convolutional neural network model.

[0068] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0069] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0070] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting molecular properties based on graph convolutional neural networks, characterized in that, include: Based on the molecular graph identifier of the target molecule, construct the target molecular graph with atoms as nodes and the target substructure graph with molecular substructure as nodes; Extract target molecular map features from the target molecular map, and extract target molecular substructure features from the target substructure map; By fusing the target molecule map features and the target molecule substructure features, the overall features of the target molecule are obtained; The overall features of the target molecule are input into a graph convolutional neural network model to obtain the molecular properties of the target molecule output by the graph convolutional neural network model.

2. The molecular property prediction method based on graph convolutional neural networks according to claim 1, characterized in that, The molecular graph identification based on the target molecule, constructing a target molecular graph with atoms as nodes and a target substructure graph with molecular substructures as nodes, includes: Based on the molecular graph identifier, the target molecular graph is constructed with atoms as nodes and chemical bonds as edges; Based on the molecular graph identifier, the target substructure graph is constructed with molecular substructures as nodes and atoms as edges.

3. The molecular property prediction method based on graph convolutional neural networks according to claim 1, characterized in that, The step of extracting target molecular map features from the target molecular map and extracting target molecular substructure features from the target substructure map includes: Based on the feature extraction model, target molecular graph features are extracted from the target molecular graph, and target molecular substructure features are extracted from the target substructure graph.

4. The molecular property prediction method based on graph convolutional neural networks according to claim 1, characterized in that, The molecular graph identifier is determined based on the simplified molecular input line representation identifier of the target molecule.

5. The molecular property prediction method based on graph convolutional neural networks according to claim 1, characterized in that, The graph convolutional neural network model is trained based on the following steps: Obtain the target datasets corresponding to different training tasks; Each target dataset is divided into a training set and a test set; Based on the training set of each target dataset, the parameters of the initial model of the graph convolutional neural network model are updated to obtain the graph convolutional neural network model.

6. The molecular property prediction method based on graph convolutional neural networks according to claim 5, characterized in that, Each target dataset includes the chemical formula, structural formula, and important physicochemical properties of the sample molecules. These important physicochemical properties include at least one of the following: boiling point, critical temperature, critical pressure, molar volume, melting point, Gibbs free energy, eccentricity factor, standard enthalpy of formation, and solubility parameter.

7. A molecular property prediction device based on graph convolutional neural networks, characterized in that, include: Construction units are used to construct target molecular graphs with atoms as nodes and target substructure graphs with molecular substructures as nodes, based on the molecular graph identifier of the target molecule. An extraction unit is used to extract target molecular map features from the target molecular map and to extract target molecular substructure features from the target substructure map; A fusion unit is used to fuse the target molecule map features and the target molecule substructure features to obtain the overall features of the target molecule. The prediction unit is used to input the overall features of the target molecule into the graph convolutional neural network model to obtain the molecular properties of the target molecule output by the graph convolutional neural network model.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the molecular property prediction method based on graph convolutional neural networks as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the molecular property prediction method based on graph convolutional neural networks as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the molecular property prediction method based on graph convolutional neural networks as described in any one of claims 1 to 6.

Images