A prediction method, device, readable storage medium and electronic equipment

Abstract

The specification discloses a prediction method, device, readable storage medium and electronic equipment. Based on a molecular structure of a to-be-predicted molecule, a specified subgraph corresponding to a substructure of the to-be-predicted molecule is determined through a graph neural network model. Then, based on specified features corresponding to the specified subgraph and preset representation features of each specified property, it is determined that the specified subgraph has the specified property. Then, the molecular property of the to-be-predicted molecule is predicted. It can be seen that the to-be-predicted molecule has its corresponding molecular property because the to-be-predicted molecule contains a substructure with a specified property. Obviously, the prediction method provides interpretability for the to-be-predicted molecule having its corresponding molecular property, and ensures the credibility of the prediction result.

Description

TECHNICAL FIELD

[0001] The present specification relates to the field of chemistry, and in particular, to a prediction method, device, readable storage medium and electronic equipment. BACKGROUND

[0002] With the development of computer technology and the need for in-depth business integration, in the medical and chemical fields, predicting molecular properties through models to perform compound screening based on the predicted molecular properties has become one of the common application scenarios of deep learning in the medical and chemical fields.

[0003] Currently, when predicting molecular properties, the molecular structure of a molecule to be predicted is usually obtained, and the molecular structure is input into a pre-trained prediction model to obtain the molecular properties output by the prediction model as the predicted molecular properties of the molecule to be predicted.

[0004] However, in general, since the prediction model can only determine the molecular properties of the molecule to be predicted, it cannot give the reason why the molecule to be predicted has its corresponding molecular properties, which leads to low credibility of the predicted molecular properties. Based on this, the present specification provides a prediction method. SUMMARY

[0005] The present specification provides a prediction method, device, readable storage medium and electronic equipment to partially solve the above problems existing in the prior art.

[0006] The present specification adopts the following technical solutions:

[0007] The present specification provides a prediction method, comprising:

[0008] establishing a molecular graph taking atoms as nodes and chemical bonds as edges according to each atom included in the molecule to be predicted and the chemical bonds between the atoms;

[0009] inputting the molecular graph into a pre-trained graph neural network model to obtain a plurality of specified subgraphs of the molecular graph output by the graph neural network model, the specified subgraph corresponding to a substructure included in the molecular structure of the molecule to be predicted;

[0010] For each specified subgraph, determining a specified feature of the specified subgraph, and determining a fusion feature according to the specified feature and a pre-stored representation feature respectively corresponding to each specified property;

[0011] inputting the fusion feature into a classification model to obtain a target classification result output by the classification model, the target classification result being used to represent a specified property possessed by a substructure corresponding to the specified subgraph;

[0012] According to the target classification results corresponding to each designated subgraph, the molecular property of the to-be-predicted molecule is predicted.

[0013] Optionally, the molecular graph is input into a pre-trained graph neural network model to obtain a plurality of designated subgraphs of the molecular graph output by the graph neural network model, specifically including:

[0014] The molecular graph is input into a pre-trained graph neural network model to obtain a confidence degree corresponding to each chemical bond in the to-be-predicted molecule output by the graph neural network, the confidence degree being used to represent a probability that an edge corresponding to the chemical bond belongs to a designated subgraph.

[0015] According to each confidence degree, each designated edge belonging to a designated subgraph is determined, and a designated subgraph is determined according to each designated edge and nodes connected to the designated edges.

[0016] Optionally, the molecular graph is input into a pre-trained graph neural network model to obtain a confidence degree corresponding to each chemical bond in the to-be-predicted molecule output by the graph neural network, specifically including:

[0017] The node features corresponding to each node in the molecular graph are determined.

[0018] For each chemical bond included in the to-be-predicted molecule, a bond feature of the chemical bond is determined according to node features of nodes connected by an edge corresponding to the chemical bond.

[0019] The bond feature is input into a pre-trained graph neural network model to obtain a confidence degree of the chemical bond output by the graph neural network.

[0020] Optionally, the node features corresponding to each node in the molecular graph are determined, specifically including:

[0021] Feature extraction is performed on each node and each edge included in the molecular graph to determine initial features corresponding to each node and initial features corresponding to each edge.

[0022] For each node in the molecular graph, neighbor nodes of the node are determined, and a node feature of the node is determined according to an initial feature of the node, initial features of the neighbor nodes, and an initial feature of an edge between the neighbor nodes and the node.

[0023] Optionally, the fusion feature is determined according to the designated feature and pre-stored representation features corresponding to each designated property, specifically including:

[0024] For each designated property, an enhanced feature of the designated property corresponding to the designated subgraph is determined according to a similarity between the designated feature and a representation feature of the designated property, and the representation feature of the designated property.

[0025] Fusing the specified features and the enhanced features of the specified sub-graphs respectively corresponding to the specified features, to obtain fused features.

[0026] Optionally, the molecular property of the to-be-predicted molecule is predicted according to the target classification results respectively corresponding to the specified sub-graphs, and specifically includes:

[0027] According to the molecular graph and the specified sub-graphs, other sub-structures in the molecular structure of the to-be-predicted molecule except for the sub-structures respectively corresponding to the specified sub-graphs are determined as specific sub-structures;

[0028] A specific sub-graph corresponding to the specific sub-structure is determined, and a specific feature of the specific sub-graph is determined;

[0029] The specific feature is input into the classification model, and a specific classification result output by the classification model is obtained;

[0030] According to the specific classification result and the target classification results respectively corresponding to the specified sub-graphs, the molecular property of the to-be-predicted molecule is predicted.

[0031] Optionally, the graph neural network model and the classification model are trained in the following manner:

[0032] For each sample molecule with a labeled specified property, a sample molecular graph with atoms as nodes and chemical bonds between the atoms as edges is established according to the atoms included in the sample molecule and the chemical bonds between the atoms, as a training sample, and the specified property is labeled as the training sample;

[0033] Each training sample is input into the to-be-trained graph neural network model, and sample specified sub-graphs respectively corresponding to the training samples output by the graph neural network are obtained;

[0034] Sample features respectively corresponding to the sample specified sub-graphs are determined, and representation features of the specified properties are determined according to the sample features and the labels respectively corresponding to the training samples;

[0035] Fused features respectively corresponding to the training samples are determined according to the sample features and the representation features of the specified properties;

[0036] Each fused feature is input into a to-be-trained classification model, and sample classification results output by the classification model are obtained;

[0037] Sample properties respectively corresponding to the training samples are determined according to the sample classification results of the sample specified sub-graphs of the training samples;

[0038] According to the sample properties and the labels of the respective training samples, the graph neural network model and the classification model are trained.

[0039] The specification provides a prediction device, comprising:

[0040] A first determination module is configured to establish a molecular graph with atoms as nodes and chemical bonds as edges according to the atoms included in the to-be-predicted molecule and the chemical bonds between the atoms;

[0041] A second determination module is configured to input the molecular graph into a pre-trained graph neural network model to obtain a plurality of specified subgraphs of the molecular graph output by the graph neural network model, the specified subgraphs corresponding to substructures included in the molecular structure of the to-be-predicted molecule;

[0042] A fusion module is configured to determine a specified feature of each specified subgraph, and determine a fusion feature according to the specified feature and a pre-stored representation feature corresponding to each specified property;

[0043] A classification module is configured to input the fusion feature into a classification model to obtain a target classification result output by the classification model, the target classification result representing a specified property possessed by the substructure corresponding to the specified subgraph;

[0044] A prediction module is configured to predict the molecular property of the to-be-predicted molecule according to the target classification result corresponding to each specified subgraph.

[0045] The specification provides a computer-readable storage medium, the storage medium storing a computer program, the computer program being executed by a processor to implement the prediction method described above.

[0046] The specification provides an electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, the processor implementing the prediction method described above when executing the program.

[0047] The above at least one technical solution adopted by the specification can achieve the following beneficial effects:

[0048] Based on the molecular structure of the to-be-predicted molecule, the specified subgraph corresponding to the substructure of the to-be-predicted molecule is determined through the graph neural network model, and then the specified property possessed by the specified subgraph is determined based on the specified feature corresponding to the specified subgraph and the representation feature of each specified property, and the molecular property of the to-be-predicted molecule is predicted.

[0049] It is evident that the molecule to be predicted possesses its corresponding molecular properties because it contains substructures with the specified properties. Clearly, this prediction method provides interpretability for the molecule to be predicted to possess its corresponding molecular properties, ensuring the reliability of the prediction results. Attached Figure Description

[0050] The accompanying drawings, which are included to provide a further understanding of this specification and form part of this specification, illustrate exemplary embodiments and their descriptions, serving to explain this specification and do not constitute an undue limitation thereof.

[0051] In the picture:

[0052] Figure 1 This is a flowchart illustrating the prediction method provided in this specification.

[0053] Figure 2 This is a schematic diagram of the molecular diagram provided in this specification;

[0054] Figure 3 This is a flowchart illustrating the prediction method provided in this specification.

[0055] Figure 4 This is a schematic diagram of the predictive device provided in this specification;

[0056] Figure 5 The corresponding information provided in this specification Figure 1 A schematic diagram of an electronic device. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of this specification clearer, the technical solutions of this specification will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this specification, and not all of them. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this specification.

[0058] The technical solutions provided in the various embodiments of this specification are described in detail below with reference to the accompanying drawings.

[0059] Figure 1 The flowchart of the prediction method provided in this specification is shown, and it specifically includes the following steps:

[0060] S100: Based on the atoms contained in the molecule to be predicted and the chemical bonds between the atoms, construct a molecular graph with atoms as nodes and chemical bonds as edges.

[0061] The embodiments of the present specification provide a prediction method, wherein the graph neural network model and the classification model involved can be obtained by pre-training. The execution process of the prediction method can be performed by an electronic device of a server for identifying the molecular properties of a to-be-predicted molecule. The electronic device performing the training process of the graph neural network model and the classification model can be the same as or different from the electronic device performing the prediction method, which is not limited in the present specification.

[0062] Unlike the current method of directly inputting the molecular structure of the to-be-predicted molecule into the prediction model, the prediction model outputs the molecular properties of the to-be-predicted molecule, but the prediction model cannot explain the molecular properties. That is, the prediction model cannot give the reason why the to-be-predicted molecule has its corresponding molecular properties. Further, it leads to the current situation that the credibility of the molecular properties of the to-be-predicted molecule obtained based on the model is low.

[0063] The present application provides a prediction method, based on the molecular structure of a to-be-predicted molecule, determines the specified subgraph corresponding to the substructure of the to-be-predicted molecule through a graph neural network model, and further predicts the molecular properties of the to-be-predicted molecule based on the specified properties of the specified subgraph. It can be seen that the prediction method in the present application can predict the molecular properties of the to-be-predicted molecule based on the properties of the substructure contained in the molecular structure of the to-be-predicted molecule. That is, the to-be-predicted molecule has its corresponding molecular properties because it contains a substructure with specified properties. Obviously, the prediction method provides explainability for the to-be-predicted molecule having its corresponding molecular properties, and ensures the credibility of the prediction result.

[0064] Based on the above brief description of the prediction method in the present application, it can be seen that the prediction method in the present specification can determine the specified subgraph through the graph neural network model. Generally, the graph neural network model is used to process the graph structure, and the graph structure can accurately represent the molecular structure of the to-be-predicted molecule. Based on this, the server can determine the molecular graph according to the molecular structure of the to-be-predicted molecule.

[0065] Specifically, for each molecule, the molecule contains atoms, and there are chemical bonds between the atoms. Among them, the chemical bond includes ionic bond and covalent bond.

[0066] Therefore, the server can determine the to-be-predicted molecule. The to-be-predicted molecule can be carried in the prediction request received by the server, or can be carried in the prediction task generated by the to-be-predicted molecule according to the preset prediction condition. Then the server can parse the received prediction request or the generated prediction task to determine the to-be-predicted molecule carried in the prediction request or the prediction task.

[0067] Then, the server can determine the atoms contained in the to-be-predicted molecule and the chemical bonds between the atoms according to the molecular structure of the to-be-predicted molecule.

[0068] Finally, the server can determine the nodes in the molecular graph according to the determined atoms, and determine the edges between the nodes in the molecular graph according to the nodes connected by each chemical bond for each chemical bond. In this way, the molecular graph with atoms as nodes and chemical bonds as edges is constructed. The constructed molecular graph contains nodes corresponding to atoms and edges between nodes. The edges between nodes in the molecular graph represent the chemical bonds between the two nodes connected by the edge. Figure 2 For example.

[0069] Figure 2 The schematic diagram of the molecular graph provided in the specification is shown in the figure. Methanol molecules are taken as an example. The server can determine the atoms contained in the methanol molecules: C, H, O, and determine the chemical bonds corresponding to each atom. Then, for each atom, determine the node corresponding to the atom, and for each chemical bond, determine the edges between the nodes in the molecular graph according to the nodes connected by the chemical bond.

[0070] S102: input the molecular graph into the pre-trained graph neural network model to obtain a plurality of specified subgraphs of the molecular graph output by the graph neural network model, the specified subgraph corresponding to a substructure contained in the molecular structure of the to-be-predicted molecule.

[0071] In one or more embodiments provided in the present application, as described above, the server can determine the specified subgraph corresponding to the substructure contained in the molecular structure of the to-be-predicted molecule through the graph neural network model, so as to predict the specified property of the to-be-predicted molecule based on the specified property possessed by the specified subgraph subsequently.

[0072] Specifically, the server is provided with a pre-trained graph neural network model. The graph neural network model is used to determine the specified subgraph contained in the molecular graph corresponding to the to-be-predicted molecule. The specified subgraph corresponds to a substructure contained in the molecular structure of the to-be-predicted molecule. The specified subgraph is a subgraph corresponding to a substructure used to represent the molecular property of the to-be-predicted molecule.

[0073] For example, the NaOH molecule, the corresponding substructure of the specified subgraph can be OH, and the property of the NaOH molecule can be predicted through the specified property of OH. Therefore, OH can be used as a substructure used to represent the molecular property of the NaOH molecule. The specified subgraph corresponding to OH can be a subgraph corresponding to a substructure used to represent the molecular property of the NaOH molecule.

[0074] Thus, the server can input the molecular graph determined in step S100 into the pre-trained graph neural network model to obtain a plurality of specified subgraphs contained in the molecular graph output by the graph neural network model. For each specified subgraph, the specified subgraph corresponds to a substructure contained in the molecular structure of the to-be-predicted molecule.

[0075] S104: For each specified subgraph, determine the specified feature of the specified subgraph, and determine the fusion feature according to the specified feature and the representation feature corresponding to each specified property pre-stored respectively.

[0076] In one or more embodiments provided in the present application, after the server determines the specified subgraph, the server can predict the molecular property corresponding to the to-be-predicted molecule based on the specified property possessed by the substructure corresponding to the specified subgraph. However, if the specified property possessed by the substructure corresponding to the specified subgraph is directly determined by classifying the specified feature corresponding to the specified subgraph, it may not be possible to explain why the substructure has the specified property. Therefore, the server can determine the specified property possessed by the substructure corresponding to the specified subgraph in the following manner: the representation feature corresponding to each specified property pre-stored respectively and the specified feature of the specified subgraph are fused to obtain a fusion result, and the specified property possessed by the substructure corresponding to the specified subgraph is determined based on the fusion result.

[0077] Specifically, the server pre-stores the representation vector corresponding to each specified property. The specified property can be a property such as toxic, non-toxic, etc., or a property such as easily soluble in water, slightly soluble in water, insoluble in water, etc. The type of the specified property can be set as needed, and the present specification does not limit this.

[0078] Meanwhile, for each specified property, the representation vector corresponding to the specified property can be used to represent the specified property. That is, the more similar the specified feature of the specified subgraph is to the representation vector of the specified property, the higher the probability that the substructure corresponding to the specified subgraph has the specified property. The atoms and the chemical bonds between the atoms contained in the specified subgraph determine the specified feature corresponding to the specified subgraph.

[0079] Finally, the server can splice the representation vector corresponding to each specified property pre-stored respectively and the specified feature corresponding to the specified subgraph, and use the splicing result as the fusion feature.

[0080] S106: Input the fusion feature into a classification model to obtain a target classification result output by the classification model, wherein the target classification result is used to represent the specified property possessed by the substructure corresponding to the specified subgraph.

[0081] In one or more embodiments provided in the present application, after determining the fusion feature, the server can classify the fusion feature through a classification model to determine that the specified substructure corresponding to the specified subgraph has a specified property.

[0082] Specifically, the server is pre-configured with a classification model for determining that the specified substructure corresponding to the specified subgraph has a specified property.

[0083] Therefore, the server can input the fusion feature into the pre-trained classification model to obtain a target classification result of the specified subgraph output by the classification model. The target classification result is the specified property of the specified substructure corresponding to the specified subgraph. Taking the specified property as toxic and non-toxic as an example, the target classification result can be that the specified substructure corresponding to the specified subgraph is toxic, or the specified substructure corresponding to the specified subgraph is non-toxic.

[0084] Of course, the target classification result can also be the probability of the specified substructure corresponding to the specified subgraph having each specified property. Similarly, taking the specified property as toxic and non-toxic as an example, the target classification result can be that the probability of the specified substructure corresponding to the specified subgraph being toxic is 20%, and the probability of the specified substructure corresponding to the specified subgraph being non-toxic is 80%, etc. The specific form of the target classification result can be set as needed, and the present specification does not limit this.

[0085] S108: predicting the molecular property of the to-be-predicted molecule according to the target classification result corresponding to each specified subgraph.

[0086] In one or more embodiments provided in the present application, the prediction method needs to predict the molecular property of the to-be-predicted molecule based on the specified property of the specified subgraph. The molecular property can be the specified property of the to-be-predicted molecule, or the probability of the to-be-predicted molecule having the specified property.

[0087] Specifically, for each specified subgraph, the server can determine the target classification result corresponding to the specified subgraph.

[0088] Then, for each specified property, the server can determine the number of specified subgraphs having the specified property as a specified number, and determine whether the specified number exceeds one half of the number of specified subgraphs contained in the to-be-predicted molecule.

[0089] If yes, the server can determine that the to-be-predicted molecule has the specified property.

[0090] If no, the server can determine that the to-be-predicted molecule does not have the specified property.

[0091] Further, if a molecule contains a toxic substructure, the molecule is likely to be toxic. If the molecule contains three specified subgraphs, and only one of the specified subgraphs corresponds to a toxic substructure, it is clear that the determined molecular properties of the to-be-predicted molecule can not include the specified property of "toxic". In order to avoid the above situation, the server can determine, for each specified property, whether the to-be-predicted molecule contains a specified subgraph with the specified property.

[0092] If yes, the server can determine that the to-be-predicted molecule has the specified property.

[0093] If no, the server can determine that the to-be-predicted molecule does not have the specified property.

[0094] Of course, the server can also determine, for each specified property, the probability that the to-be-predicted molecule has the specified property based on the probabilities that each specified subgraph contained by the to-be-predicted molecule has the specified property.

[0095] Therefore, the server can determine the molecular properties of the to-be-predicted molecule based on the determined specified properties of the to-be-predicted molecule or the probabilities that the to-be-predicted molecule has the specified properties. The server can directly use the probabilities that the to-be-predicted molecule has the specified properties as the molecular properties of the to-be-predicted molecule, or can use the specified properties with probabilities exceeding a specific probability as the molecular properties of the to-be-predicted molecule. How to determine the molecular properties of the to-be-predicted molecule based on the target classification results of each specified subgraph can be set as needed, and the present specification does not limit this.

[0096] Based on Figure 1 As shown in the prediction method, based on the molecular structure of the to-be-predicted molecule, the specified subgraph corresponding to the substructure of the to-be-predicted molecule is determined through the graph neural network model, and then the molecular properties of the to-be-predicted molecule are predicted based on the specified properties of the specified subgraph. It can be seen that the to-be-predicted molecule has its corresponding molecular properties because the to-be-predicted molecule contains a substructure with a specified property. Obviously, the prediction method provides interpretability for the to-be-predicted molecule having its corresponding molecular properties, and ensures the credibility of the prediction result.

[0097] Further, for a molecule, the chemical bonds possessed by the molecule can generally affect the physical properties of the molecule. The prediction method provided in the present application aims to determine the specified subgraph that can be used to characterize the molecular properties of the to-be-predicted molecule, and then predict the molecular properties of the to-be-predicted molecule by the specified properties possessed by the specified subgraph. Based on the same idea, if the chemical bond that can be used to characterize the molecular properties of the to-be-predicted molecule is determined, and then the specified subgraph is determined based on the determined chemical bond, the molecular properties of the to-be-predicted molecule can be determined based on the target classification result of the subgraph corresponding to the specified subgraph.

[0098] Specifically, the server can input the molecular graph determined in step S100 into the pre-trained graph neural network model to obtain the confidence corresponding to each chemical bond in the to-be-predicted molecule output by the graph neural network model. The graph neural network is used to determine the specified subgraph contained in the molecular graph corresponding to the to-be-predicted molecule. For each chemical bond, the confidence corresponding to the chemical bond is used to represent the probability that the edge corresponding to the chemical bond belongs to the specified subgraph.

[0099] Therefore, the server can determine the specified edges belonging to the specified subgraph according to the determined confidences.

[0100] Finally, the server can determine the specified subgraph according to the determined specified edges and the nodes connected to the specified edges.

[0101] Further, for each chemical bond, the properties of the chemical bond can be represented based on the properties of the atoms connected to the chemical bond. Therefore, when determining the confidence corresponding to each chemical bond, the server can also determine the bond feature corresponding to the chemical bond for each chemical bond, and then determine the confidence corresponding to the chemical bond based on the bond feature.

[0102] Specifically, the server can determine the node feature corresponding to each node in the molecular graph.

[0103] Therefore, for each chemical bond contained in the to-be-predicted molecule, the server can determine the two nodes connected by the edge corresponding to the chemical bond, and determine the node features of the two nodes by performing feature extraction on the two nodes, respectively.

[0104] Then, the server can concatenate the node features of the two nodes as the bond feature of the chemical bond.

[0105] Finally, the server can input the bond feature into the pre-trained graph neural network model to obtain the confidence of the chemical bond output by the graph neural network.

[0106] Of course, the server can also extract features of the chemical bond, determine the initial features corresponding to the chemical bond, and then fuse the initial features corresponding to the chemical bond and the node features of the two nodes, and take the fusion result as the bond feature of the chemical bond.

[0107] In addition, in the present specification, for each atom in the molecular graph, the properties corresponding to the atom are not only affected by the atom itself, but also affected by other atoms connected to the atom. Therefore, when determining the node features of the nodes corresponding to the respective atoms, the features of the neighbor nodes of the node can also be determined based on the features of the neighbor nodes.

[0108] Specifically, for each node and each edge included in the molecular graph, the server can extract features of the node and the edge, and determine the initial features of the node and the initial features of the edge.

[0109] Then, for each node in the molecular graph, the server can determine the neighbor nodes of the node.

[0110] Finally, the server can determine the node features of the node according to the initial features of the node, the initial features of the neighbor nodes, and the initial features of the edges between the neighbor nodes and the node.

[0111] Of course, it should be noted that the server can also take the determined node features of the node as the initial features of the node again, and continue to determine the node features of the node based on the initial features of the node and the initial features of the neighbor nodes of the node according to the re-determined initial features of the node. To achieve the purpose of transmitting the properties of the atoms in the molecular graph along the chemical bonds, and further ensure the accuracy of the confidence of the chemical bond determined based on the node features of the node.

[0112] Further, in the present application, the purpose of determining the fused features is to determine the specified properties possessed by the specified features based on the similarity between the specified features and the representation features respectively corresponding to the pre-stored specified properties. For each specified property, if the specified features are enhanced based on the similarity between the representation features corresponding to the specified property and the specified features, the accuracy of the target classification result determined based on the enhancement result will be higher. Based on this, the server can enhance the specified features based on the representation features of the specified properties.

[0113] Specifically, the server can determine, for each specified property, the enhanced features corresponding to the specified property in the specified subgraph according to the similarity between the representation features of the specified property and the specified features of the specified subgraph, and the representation features of the specified property.

[0114] After determining the specified properties respectively correspond to the enhanced features of the specified sub-graphs, the server can fuse the specified features and the enhanced features of the specified properties respectively corresponding to the specified sub-graphs to obtain fused features. Then, the target classification results of the specified sub-graphs can be obtained based on the fused features.

[0115] For each specified property, the representation feature of the specified property can be determined in the following manner:

[0116] For each specified property, the molecular graphs of the samples respectively corresponding to the specified property are input into the pre-trained graph neural network model to obtain the specified sub-graphs respectively corresponding to the samples of the specified property output by the graph neural network model. Then, the representation feature of the specified property is determined based on the specified features respectively corresponding to the specified sub-graphs.

[0117] Of course, the server can also select any sample molecule from the samples respectively corresponding to the specified property, input the molecular graph of the sample molecule into the pre-trained graph neural network model to obtain the specified sub-graph corresponding to the sample molecule output by the graph neural network model, and then take the specified feature of the specified sub-graph as the representation feature of the specified property.

[0118] Specifically, how to determine the representation features respectively corresponding to the specified properties can be set as needed, and the present specification does not limit this.

[0119] Further, for the to-be-predicted molecule, in addition to the sub-structures corresponding to the specified sub-graphs, the molecular structure of the to-be-predicted molecule also contains other sub-structures, and the properties of the other sub-structures can also affect the properties of the to-be-predicted molecule. Therefore, the server can also take the other sub-structures as specific sub-structures, and determine the properties of the to-be-predicted molecule based on the specific sub-structures.

[0120] Specifically, the server can determine the other sub-structures in the molecular structure of the to-be-predicted molecule except the sub-structures corresponding to the specified sub-graphs as specific sub-structures based on the molecular graph and the specified graphs.

[0121] Therefore, the server can determine the specific sub-graph corresponding to the specific sub-structure, and determine the specific feature of the specific sub-graph.

[0122] Then, the server can input the specific feature into the classification model to obtain the specific classification result output by the classification model.

[0123] Finally, the server can predict the molecular properties of the to-be-predicted molecule based on the specific classification result and the target classification results respectively corresponding to the specified sub-graphs. For example, Figure 3 as shown.

[0124] Figure 3A flowchart of a prediction method provided in the specification is shown, wherein the server can input a molecular graph of a to-be-predicted molecule into a graph neural network model to obtain a specified subgraph and a specific subgraph output by the graph neural network model. Then, the server can fuse a specified feature of the specified subgraph with a representation feature of each specified property in a preset manner to obtain a fused feature, and then input the fused feature and the specific subgraph into a classification model respectively to obtain a target classification result corresponding to the fused feature and a specific classification result corresponding to the specific subgraph output by the classification model. Finally, the server can determine a prediction result based on the specific classification result and the target classification result, and the prediction result is a molecular property possessed by the to-be-predicted molecule.

[0125] In addition, the graph neural network model and the classification model in the specification can be trained in the following manner:

[0126] Specifically, the server can obtain a plurality of sample molecules with a labeled specified property. For each sample molecule with a labeled specified property, a sample molecular graph with atoms as nodes and chemical bonds as edges is established according to the atoms included in the sample molecule and the chemical bonds between the atoms, as a training sample. At the same time, the server can take the specified property as the label of the training sample.

[0127] Secondly, the server can input each training sample into a to-be-trained graph neural network model to obtain a sample specified subgraph corresponding to each training sample output by the graph neural network.

[0128] Then, the server can determine a sample feature corresponding to each sample specified subgraph, and determine a representation feature of each specified property according to each sample feature and the label corresponding to each training sample.

[0129] Then, the server can determine a fused feature corresponding to each training sample according to each sample feature and the representation feature of each specified property.

[0130] After that, the server can input each fused feature into a to-be-trained classification model to obtain a sample classification result output by the classification model.

[0131] Therefore, the server can determine a sample property corresponding to each training sample according to the sample classification result of the sample specified subgraph corresponding to each training sample.

[0132] Finally, the server can determine a loss according to the sample property corresponding to each training sample and the label thereof, and train the graph neural network model and the classification model with the goal of minimizing the loss.

[0133] Further, the graph neural network in the specification can also be trained in the following manner:

[0134] Specifically, the server can obtain a plurality of sample molecules with a specified property labeled. And for each sample molecule with a specified property labeled, a sample molecule graph with atoms as nodes and chemical bonds as edges is established according to each atom contained in the sample molecule and the chemical bonds between the atoms, as a training sample.

[0135] Secondly, the server can determine, for each training sample, a specified subgraph contained in the training sample as a target sample subgraph of the training sample.

[0136] Then, the server can input each training sample into a graph neural network model to be trained respectively, to obtain a sample specified subgraph corresponding to each training sample respectively output by the graph neural network.

[0137] Finally, the server can determine a loss of the graph neural network model according to the sample specified subgraph and the target sample subgraph corresponding to each training sample respectively, and adjust the model parameters of the graph neural network model with the loss minimum as the optimization goal.

[0138] Based on the same idea, the present specification also provides a prediction device, as shown in Figure 4 .

[0139] Figure 4 The prediction device provided in the present specification, wherein:

[0140] The first determination module 200 is configured to establish a molecule graph with atoms as nodes and chemical bonds as edges according to each atom contained in the obtained to-be-predicted molecule and the chemical bonds between the atoms.

[0141] The second determination module 202 is configured to input the molecule graph into a pre-trained graph neural network model to obtain a plurality of specified subgraphs of the molecule graph output by the graph neural network model, the specified subgraph corresponding to a substructure contained in the molecular structure of the to-be-predicted molecule.

[0142] The fusion module 204 is configured to determine a specified feature of each specified subgraph, and determine a fusion feature according to the specified feature and the representation feature corresponding to each specified property respectively pre-stored.

[0143] The classification module 206 is configured to input the fusion feature into a classification model to obtain a target classification result output by the classification model, the target classification result being used to represent that the specified subgraph corresponding to the substructure has a specified property.

[0144] The prediction module 208 is configured to predict the molecular property of the to-be-predicted molecule according to the target classification result corresponding to each specified subgraph respectively.

[0145] Optionally, the first determining module 200 is configured to input the molecular graph into a pre-trained graph neural network model to obtain confidence degrees corresponding to respective chemical bonds in the to-be-predicted molecule output by the graph neural network, the confidence degree being used to represent a probability that an edge corresponding to the chemical bond belongs to a specified subgraph, determine respective specified edges belonging to the specified subgraph according to the respective confidence degrees, and determine the specified subgraph according to the respective specified edges and nodes connected to the respective specified edges.

[0146] Optionally, the first determining module 200 is configured to determine node features corresponding to respective nodes in the molecular graph, determine a bond feature of each chemical bond included in the to-be-predicted molecule according to node features of nodes connected to an edge corresponding to the chemical bond, and input the bond feature into a pre-trained graph neural network model to obtain a confidence degree of the chemical bond output by the graph neural network.

[0147] Optionally, the first determining module 200 is configured to perform feature extraction on respective nodes and edges included in the molecular graph to determine initial features corresponding to the respective nodes and initial features corresponding to the respective edges, determine neighbor nodes of each node in the molecular graph, and determine a node feature of the node according to an initial feature of the node, initial features of the neighbor nodes, and an initial feature of an edge between the neighbor nodes and the node.

[0148] Optionally, the fusion module 204 is configured to determine, for each specified property, an enhanced feature corresponding to the specified property according to a similarity between the specified feature and a representation feature of the specified property and the representation feature of the specified property, and fuse the specified feature and the enhanced feature corresponding to the specified property of each specified property to obtain a fusion feature.

[0149] Optionally, the prediction module 208 is configured to determine, according to the molecular graph and the respective specified subgraphs, other substructures in a molecular structure of the to-be-predicted molecule except for respective substructures corresponding to the respective specified subgraphs as specific substructures, determine a specific subgraph corresponding to the specific substructures, determine a specific feature of the specific subgraph, input the specific feature into the classification model to obtain a specific classification result output by the classification model, and predict a molecular property of the to-be-predicted molecule according to the specific classification result and target classification results respectively corresponding to the respective specified subgraphs.

[0150] The apparatus further includes:

[0151] The training module 210 is configured to train the graph neural network model and the classification model in the following manner: for each sample molecule with a labeled specified property, a sample molecule graph with atoms as nodes and chemical bonds as edges is established according to the atoms included in the sample molecule and the chemical bonds between the atoms, as a training sample, and the specified property is taken as the label of the training sample; each training sample is input into the graph neural network to be trained, to obtain a sample specified subgraph corresponding to each training sample output by the graph neural network; sample features corresponding to each sample specified subgraph are determined; representation features of each specified property are determined according to the sample features and the labels corresponding to each training sample; fusion features corresponding to each training sample are determined according to the sample features and the representation features of each specified property; each fusion feature is input into the classification model to be trained, to obtain a sample classification result output by the classification model; sample properties corresponding to each training sample are determined according to the sample classification results of the sample specified subgraphs of each training sample; and the graph neural network model and the classification model are trained according to the sample properties corresponding to each training sample and the labels thereof.

[0152] The present specification also provides a computer-readable storage medium storing a computer program, which can be used to execute the above Figure 1 The prediction method is provided.

[0153] The present specification also provides a computer program product, which can be used to execute the above Figure 5 The schematic structural diagram of the electronic device is shown. As Figure 5 described above, at the hardware level, the electronic device includes a processor, an internal bus, a network interface, a memory and a non-volatile memory, and of course, other hardware required by a business. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs to implement the prediction method described above Figure 1 . Of course, in addition to the software implementation, the present specification does not exclude other implementation manners, such as a logic device or a combination of software and hardware, and so on, that is, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or a logic device.

[0154] In the 1990s, it was relatively easy to distinguish whether an improvement in a technology was a hardware improvement (e.g., an improvement in the circuit structure of a diode, transistor, switch, etc.) or a software improvement (an improvement in a method flow). However, as technology has evolved, many improvements in method flows today can be considered as direct improvements in hardware circuit structures. Designers almost always obtain the corresponding hardware circuit structures by programming the improved method flows into hardware circuits. Therefore, it cannot be said that an improvement in a method flow cannot be implemented using hardware entity modules. For example, a programmable logic device (PLD) (e.g., a field programmable gate array (FPGA)) is an integrated circuit whose logic function is determined by user programming of the device. A digital system is "integrated" on a PLD by the designer programming the PLD, rather than by ordering a chip manufacturer to design and fabricate a custom integrated circuit chip. Moreover, instead of manually fabricating integrated circuit chips, this programming is now mostly implemented using "logic compiler" software, which is similar to software compilers used in program development, and the original code to be compiled is written in a specific programming language, which is called a hardware description language (HDL), and there are many such languages, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc., and the most commonly used are VHDL (Very-High-Speed Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should be aware that, as long as the method flow is logically programmed in the above-mentioned hardware description languages and programmed into an integrated circuit, a hardware circuit implementing the logical method flow can be easily obtained.

[0155] The controller can be implemented in any suitable way, for example, the controller can take the form of a microprocessor or processor and a computer readable medium storing computer readable program code, such as software or firmware, executable by the (micro)processor, logic gates, switches, an application specific integrated circuit (ASIC), a programmable logic controller and an embedded microcontroller, examples of which include but are not limited to the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20 and Silicone Labs C8051F320, the memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art will also know that, in addition to being implemented in pure computer readable program code, the controller can equally well be implemented to perform the same functions using logic gates, switches, an application specific integrated circuit, a programmable logic controller and an embedded microcontroller, etc. by means of a logical programming of the method steps. The controller can thus be considered as a hardware component, and the means comprised therein for performing the various functions can be considered as structures within the hardware component. Alternatively, the means for performing the various functions can even be considered as both a software module implementing the method and a structure within the hardware component.

[0156] The systems, apparatuses, modules or units illustrated by the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer can be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0157] For the sake of description, the above apparatuses are described in various units by functions respectively. Of course, the functions of the units can be implemented in one or more software and / or hardware in the implementation of the present specification.

[0158] Those skilled in the art will understand that the embodiments of the present specification can be provided as a method, a system or a computer program product. Therefore, the present specification can take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware aspects. Moreover, the present specification can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0159] The specification is presented with reference to flow diagrams and / or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the specification. It will be understood that each block of the flow diagrams and / or block diagrams, and combinations of blocks in the flow diagrams and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded processing device or other programmable device (s) to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable device, create means for implementing the functions specified in the flow diagrams and / or block diagrams block or blocks. Figure 1 The flow diagrams and / or block diagrams in this specification can present a flow diagram and / or block diagram illustrating the functionality and / or operations of a possible implementation of methods, apparatus (systems) and computer program products according to this specification. Flow diagrams and / or block diagrams should be understood as merely representational and each block of the flow diagrams and / or block diagrams and combinations of blocks in the flow diagrams and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded processing device or other programmable device (s) to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable device, create means for implementing the functions specified in the flow diagrams and / or block diagrams block or blocks. Figure 1 The flow diagrams and / or block diagrams in this specification can present a flow diagram and / or block diagram illustrating the functionality and / or operations of a possible implementation of methods, apparatus (systems) and computer program products according to this specification. Flow diagrams and / or block diagrams should be understood as merely representational and each block of the flow diagrams and / or block diagrams and combinations of blocks in the flow diagrams and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded processing device or other programmable device (s) to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable device, create means for implementing the functions specified in the flow diagrams and / or block diagrams block or blocks.

[0160] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable device to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instructions which implement the function specified in the flow diagrams and / or block diagrams block or blocks. Figure 1 The flow diagrams and / or block diagrams in this specification can present a flow diagram and / or block diagram illustrating the functionality and / or operations of a possible implementation of methods, apparatus (systems) and computer program products according to this specification. Flow diagrams and / or block diagrams should be understood as merely representational and each block of the flow diagrams and / or block diagrams and combinations of blocks in the flow diagrams and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded processing device or other programmable device (s) to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable device, create means for implementing the functions specified in the flow diagrams and / or block diagrams block or blocks. Figure 1 The flow diagrams and / or block diagrams in this specification can present a flow diagram and / or block diagram illustrating the functionality and / or operations of a possible implementation of methods, apparatus (systems) and computer program products according to this specification. Flow diagrams and / or block diagrams should be understood as merely representational and each block of the flow diagrams and / or block diagrams and combinations of blocks in the flow diagrams and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded processing device or other programmable device (s) to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable device, create means for implementing the functions specified in the flow diagrams and / or block diagrams block or blocks.

[0161] These computer program instructions can also be loaded onto a computer or other programmable device to cause a series of operational steps to be performed on the computer or other programmable device to produce a computer implemented process such that the instructions which execute on the computer or other programmable device provide steps for implementing the functions specified in the flow diagrams and / or block diagrams block or blocks. Figure 1 The flow diagrams and / or block diagrams in this specification can present a flow diagram and / or block diagram illustrating the functionality and / or operations of a possible implementation of methods, apparatus (systems) and computer program products according to this specification. Flow diagrams and / or block diagrams should be understood as merely representational and each block of the flow diagrams and / or block diagrams and combinations of blocks in the flow diagrams and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded processing device or other programmable device (s) to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable device, create means for implementing the functions specified in the flow diagrams and / or block diagrams block or blocks. ​ The flow diagrams and / or block diagrams in this specification can present a flow diagram and / or block diagram illustrating the functionality and / or operations of a possible implementation of methods, apparatus (systems) and computer program products according to this specification. Flow diagrams and / or block diagrams should be understood as merely representational and each block of the flow diagrams and / or block diagrams and combinations of blocks in the flow diagrams and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded processing device or other programmable device (s) to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable device, create means for implementing the functions specified in the flow diagrams and / or block diagrams block or blocks.

[0162] In a typical configuration, a computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0163] The memory can include non-persistent memory and / or volatile memory, such as random access memory (RAM) and / or cache memory, non-volatile memory, such as read-only memory (ROM), EPROM, and / or flash memory, etc. The memory is an example of computer readable media.

[0164] Computer-readable media includes permanent and non-permanent, movable and non-movable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible to a computing device. According to the definition herein, computer-readable media does not include transitory media such as modulated data signals and carriers.

[0165] It should also be noted that the terms "comprising", "containing", or any other variant thereof are intended to cover non-exclusive inclusion, such that a process, method, article or apparatus that comprises a list of elements does not only include those elements, but can also include other elements not expressly listed or inherent to such process, method, article or apparatus. Without more limitations, the element defined by the statement "comprising a" does not exclude the presence of additional identical elements in the process, method, article or apparatus that includes the element.

[0166] Those skilled in the art will appreciate that embodiments of the present specification can be provided as methods, systems or computer program products. Therefore, the present specification can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Moreover, the present specification can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0167] The present specification can be described in the general context of computer-executable instructions, such as program modules, executed by computers. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. The present specification can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in both local and remote computer storage media including storage devices.

[0168] The various embodiments described in this specification are described using a numbering of embodiments approach: these are each individually integrated contributions pertaining to different aspects of the description. For each embodiment, the description focuses on the differences from the other embodiments. Each embodiment is to be read in isolation, with the understanding that the same or similar features from other embodiments can be combined with the features of the respective embodiment. In particular, the description of the system embodiments is kept relatively short, as the system embodiments are largely analogous to the method embodiments.

[0169] The above only describes the embodiments of the present specification and is not intended to limit the present specification. The present specification can have various modifications and changes for those skilled in the art. Any modification, equivalent replacement, improvement, etc. within the spirit and principle of the present specification shall be included in the scope of claims of the present specification.

Claims

1. A method of prediction, characterized by, The method comprises: According to the obtained atoms contained in the to-be-predicted molecule and the chemical bonds between the atoms, a molecular graph is established, taking atoms as nodes and chemical bonds as edges; input the molecular graph into a pre-trained graph neural network model to obtain a plurality of specified subgraphs of the molecular graph output by the graph neural network model, the specified subgraphs corresponding to the substructures contained in the molecular structure of the to-be-predicted molecule; For each specified subgraph, determine the specified feature of the specified subgraph, and determine the fusion feature according to the specified feature and the pre-stored representation feature corresponding to each specified property, respectively; input the fusion feature into a classification model to obtain a target classification result output by the classification model, the target classification result being used to represent that the specified subgraph corresponding to the specified property has a specified property; According to the target classification result corresponding to each specified subgraph, the molecular property of the to-be-predicted molecule is predicted; According to the specified feature and the pre-stored representation feature corresponding to each specified property, respectively, the fusion feature is determined, specifically including: For each specified property, according to the similarity between the specified feature and the representation feature of the specified property, and the representation feature of the specified property, the enhanced feature of the specified property corresponding to the specified subgraph is determined; fuse the specified feature and the enhanced feature of each specified property corresponding to the specified subgraph to obtain the fusion feature.

2. The method of claim 1, wherein, input the molecular graph into a pre-trained graph neural network model to obtain a plurality of specified subgraphs of the molecular graph output by the graph neural network model, specifically including: input the molecular graph into a pre-trained graph neural network model to obtain a plurality of specified subgraphs of the molecular graph output by the graph neural network model, specifically including: According to each confidence, determine the specified edge belonging to the specified subgraph, and according to each specified edge and the node connected to the specified edge, determine the specified subgraph.

3. The method of claim 2, wherein, input the molecular graph into a pre-trained graph neural network model to obtain a plurality of specified subgraphs of the molecular graph output by the graph neural network model, specifically including: determine the node feature corresponding to each node in the molecular graph; For each chemical bond contained in the to-be-predicted molecule, determine the bond feature of the chemical bond according to the node feature of the node connected by the edge corresponding to the chemical bond; input the bond feature into a pre-trained graph neural network model to obtain the confidence of the chemical bond output by the graph neural network.

4. The method of claim 3, wherein, determine the node feature corresponding to each node in the molecular graph, specifically including: extract features from each node and each edge contained in the molecular graph to determine the initial feature corresponding to each node and the initial feature corresponding to each edge; For each node in the molecular graph, determine the neighbor node of the node, and determine the node feature of the node according to the initial feature of the node, the initial feature of the neighbor node, and the initial feature of the edge between the neighbor node and the node.

5. The method of claim 1, wherein, According to the target classification results corresponding to the specified subgraphs respectively, the molecular property of the to-be-predicted molecule is predicted, and the prediction specifically comprises: According to the molecular graph and each specified subgraph, other substructures in the molecular structure of the to-be-predicted molecule except for each substructure corresponding to each specified subgraph are determined as specific substructures; The specific subgraph corresponding to the specific substructure is determined, and a specific feature of the specific subgraph is determined; The specific feature is input into the classification model to obtain a specific classification result output by the classification model; According to the specific classification result and the target classification results corresponding to each specified subgraph respectively, the molecular property of the to-be-predicted molecule is predicted.

6. The method of claim 1, wherein, The graph neural network model and the classification model are trained in the following manner: For each sample molecule with a labeled specified property, a sample molecular graph with atoms as nodes and chemical bonds as edges is established according to each atom included in the sample molecule and the chemical bonds between the atoms, as a training sample, and the specified property is taken as a label of the training sample; Each training sample is input into the to-be-trained graph neural network model to obtain a sample specified subgraph corresponding to each training sample output by the graph neural network; Sample features corresponding to each sample specified subgraph are determined, and representation features of each specified property are determined according to each sample feature and the label of each training sample; According to each sample feature and the representation features of each specified property, a fusion feature corresponding to each training sample is determined; Each fusion feature is input into the to-be-trained classification model to obtain a sample classification result output by the classification model; According to the sample classification result of each sample specified subgraph of each training sample, a sample property corresponding to each training sample is determined; The graph neural network model and the classification model are trained according to the sample properties corresponding to each training sample and the labels thereof.

7. A prediction device characterized by comprising: The device comprises: A first determination module configured to establish a molecular graph with atoms as nodes and chemical bonds as edges according to each atom included in the to-be-predicted molecule and the chemical bonds between the atoms; A second determination module configured to input the molecular graph into a pre-trained graph neural network model to obtain a plurality of specified subgraphs of the molecular graph output by the graph neural network model, the specified subgraphs corresponding to substructures included in the molecular structure of the to-be-predicted molecule; A fusion module configured to determine, for each specified subgraph, a specified feature of the specified subgraph, and determine a fusion feature according to the specified feature and representation features corresponding to each specified property respectively; A classification module configured to input the fusion feature into a classification model to obtain a target classification result output by the classification model, the target classification result being used to represent a specified property possessed by a substructure corresponding to the specified subgraph; A prediction module configured to predict the molecular property of the to-be-predicted molecule according to the target classification results corresponding to each specified subgraph respectively. The fusion module is specifically configured to determine a specified feature of each specified subgraph; for each specified property, determine an enhanced feature of the specified property corresponding to the specified subgraph according to a similarity between the specified feature and a representation feature of the specified property and the representation feature of the specified property; and fuse the specified feature and the enhanced feature of each specified property corresponding to the specified subgraph to obtain a fused feature.

8. A computer readable storage medium, the storage medium having stored thereon a computer program, characterized in that, The computer program, when executed by a processor, implements the method of any one of claims 1-6.

9. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor, when executing the program, implements the method of any one of claims 1-6.

Images