WO2024212719A1 - Prediction method and apparatus, readable storage medium and electronic device - Google Patents

Abstract

A prediction method and apparatus, a readable storage medium and an electronic device. The method comprises: on the basis of the molecular structure of a molecule to be predicted, by means of a graph neural network model, determining specified sub-graphs corresponding to sub-structures of said molecule, so as to determine specified properties of the specified sub-graphs on the basis of specified features corresponding to the specified sub-graphs and preset characterization features of each specified property, thus predicting molecular properties of said molecule. Therefore, said molecule has corresponding molecular properties since said molecule comprises sub-structures having specified properties. Obviously, the prediction method provides explainability for said molecule to have the corresponding molecular properties, thus ensuring the credibility of the prediction result.

Description

Prediction method, device, readable storage medium and electronic device Technical Field

The present disclosure relates to the field of chemistry, and in particular to a prediction method, device, readable storage medium and electronic device.

Background Art

With the development of computer technology and the need for in-depth business integration, in the fields of medicine and chemistry, predicting molecular properties through models and screening compounds based on the predicted molecular properties has become one of the common application scenarios of deep learning in the fields of medicine and chemistry.

Currently, when predicting molecular properties, it is usually necessary to obtain the molecular structure of the molecule to be predicted, and then input the molecular structure 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.

However, usually, since the prediction model can only determine the molecular properties of the molecule to be predicted, but cannot provide the reason why the molecule to be predicted has its corresponding molecular properties, the credibility of the predicted molecular properties is not high.

Summary of the invention

The present disclosure provides a prediction method, device, readable storage medium and electronic device.

The present disclosure adopts the following technical solutions:

The present disclosure provides a prediction method, comprising:

According to the obtained atoms contained in the molecule to be predicted and the chemical bonds between the atoms, a molecular graph is established with the atoms as nodes and the chemical bonds as edges;

Inputting the molecular graph into a pre-trained graph neural network model to obtain several specified subgraphs of the molecular graph output by the graph neural network model, wherein the specified subgraphs correspond to substructures contained in the molecular structure of the molecule to be predicted;

For each designated sub-graph, a designated feature of the designated sub-graph is determined, and a fusion feature is determined according to the designated feature and the representation features corresponding to the pre-stored designated properties;

Inputting 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 characterize a specified property of a substructure corresponding to the specified subgraph;

The molecular properties of the molecules to be predicted are predicted according to the target classification results corresponding to the designated subgraphs.

Optionally, the molecular graph is input into a pre-trained graph neural network model to obtain several specified subgraphs of the molecular graph output by the graph neural network model, specifically including:

Inputting the molecular graph into a pre-trained graph neural network model, obtaining the confidence corresponding to each chemical bond in the molecule to be predicted output by the graph neural network model, wherein the confidence is used to characterize the probability that the edge corresponding to the chemical bond belongs to a specified subgraph;

According to each confidence level, each designated edge belonging to the designated subgraph is determined, and according to each designated edge and nodes connecting the designated edges, the designated subgraph is determined.

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

Determine the node features corresponding to each node in the molecular graph;

For each chemical bond contained in the molecule to be predicted, determining the bond feature of the chemical bond according to the node features of the nodes connected by the edge corresponding to the chemical bond;

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

Optionally, determining the node features corresponding to each node in the molecular graph specifically includes:

Extracting features from each node and each edge in the molecular graph to determine initial features corresponding to each node and initial features corresponding to each edge;

For each node in the molecular graph, the neighbor nodes of the node are determined, and the node features of the node are determined 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.

Optionally, determining the fusion feature according to the designated feature and the representation features corresponding to the pre-stored designated properties respectively includes:

For each designated property, determining, according to the similarity between the designated feature and the characterizing feature of the designated property, and the characterizing feature of the designated property, an enhanced feature of the designated property corresponding to the designated subgraph;

The designated feature and the enhanced features of each designated property corresponding to the designated sub-graph are fused to obtain the fused feature.

Optionally, predicting the molecular properties of the molecule to be predicted according to the target classification results corresponding to the designated subgraphs respectively includes:

According to the molecular graph and the designated subgraphs, determining other substructures in the molecular structure of the molecule to be predicted except for the substructures corresponding to the designated subgraphs as specific substructures;

Determining a specific subgraph corresponding to the specific substructure, and determining specific features of the specific subgraph;

Inputting the specific feature into the classification model to obtain a specific classification result output by the classification model;

The molecular properties of the molecule to be predicted are predicted according to the specific classification results and the target classification results respectively corresponding to the designated subgraphs.

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

For each sample molecule that has been labeled with a specified property, a sample molecule graph is established with each atom as a node and the chemical bonds as an edge according to the atoms contained in the sample molecule and the chemical bonds between the atoms, as a training sample, and the specified property is used as a label for the training sample;

Input each training sample into the graph neural network model to be trained, and obtain a sample specified subgraph corresponding to each training sample output by the graph neural network model to be trained;

Determine the sample features corresponding to the designated subgraphs of each sample, and determine the characterization features of each designated property according to the sample features and the labels corresponding to the training samples;

Determine, according to the features of each sample and the characterization features of each specified property, the fusion features corresponding to each training sample;

Inputting each fusion feature into the classification model to be trained to obtain a sample classification result output by the classification model to be trained;

Determine the sample properties corresponding to each training sample according to the sample classification results of the sample designated subgraphs corresponding to each training sample;

The graph neural network model and the classification model are trained according to the sample properties and annotations corresponding to each training sample.

The present disclosure provides a prediction device, comprising:

A first determination module is used to establish a molecular graph with each atom as a node and the chemical bonds as an edge according to each atom contained in the acquired molecule to be predicted and the chemical bonds between the atoms;

A second determination module is used 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, wherein the specified subgraphs correspond to substructures contained in the molecular structure of the molecule to be predicted;

A fusion module, used for determining, for each specified sub-graph, a specified feature of the specified sub-graph, and determining a fusion feature according to the specified feature and the representation features corresponding to the pre-stored specified properties;

A classification module, used for inputting 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 characterize a specified property of a substructure corresponding to the specified subgraph;

A prediction module is used to predict the target classification results of the molecules to be predicted according to the target classification results corresponding to each specified subgraph. Molecular properties.

The present disclosure provides a computer-readable storage medium, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned prediction method is implemented.

The present disclosure provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the above-mentioned prediction method is implemented when the processor executes the program.

BRIEF DESCRIPTION OF THE DRAWINGS

The drawings described herein are used to provide a further understanding of the present disclosure and constitute a part of the present disclosure. The illustrative embodiments of the present disclosure and their descriptions are used to explain the present disclosure and do not constitute an improper limitation on the present disclosure. In the drawings:

FIG1 is a schematic diagram of a flow chart of a prediction method provided by the present disclosure;

FIG2 is a schematic diagram of a molecular graph provided by the present disclosure;

FIG3 is a schematic diagram of a flow chart of a prediction method provided by the present disclosure;

FIG4 is a schematic diagram of the structure of a prediction device provided by the present disclosure;

FIG. 5 is a schematic diagram of an electronic device corresponding to FIG. 1 provided by the present disclosure.

DETAILED DESCRIPTION

In order to make the purpose, technical solutions and advantages of the present disclosure clearer, the technical solutions of the present disclosure will be clearly and completely described below in combination with the specific embodiments of the present disclosure and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present disclosure, not all of the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present disclosure.

In the related art, the molecular structure of the molecule to be predicted is directly input into the prediction model, and the prediction model outputs the molecular properties of the molecule to be predicted, but the prediction model cannot explain the molecular properties. That is, the prediction model cannot give the reason why the molecule to be predicted has its corresponding molecular properties. This leads to the low credibility of the molecular properties of the molecule to be predicted based on the current model.

The present application provides a prediction method, which is based on the molecular structure of the molecule to be predicted, and through a graph neural network model, determines the specified subgraph corresponding to the substructure of the molecule to be predicted, and then predicts the molecular properties of the molecule to be predicted 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 molecule to be predicted based on the properties of the substructure contained in the molecular structure of the molecule to be predicted, that is, the molecule to be predicted has its corresponding molecular properties because the molecule to be predicted contains a substructure with specified properties. Obviously, the prediction method provides explainability for the molecule to be predicted to have its corresponding molecular properties, and ensures the credibility of the prediction result.

In the prediction method provided in the embodiment of the present disclosure, the graph neural network model and classification model involved may be pre-trained. The execution process of the prediction method may be performed by an electronic device for identifying the molecular properties of the molecule to be predicted, such as a server. The electronic device that executes the training process of the graph neural network model and the classification model may be the same as or different from the electronic device that executes the prediction method, and the present disclosure does not limit this.

The technical solutions provided by various embodiments of the present disclosure are described in detail below in conjunction with the accompanying drawings.

FIG1 is a schematic flow chart of the prediction method provided by the present disclosure, which specifically includes the following steps S100 to S108 .

S100: Building a molecular graph with atoms as nodes and chemical bonds as edges based on the acquired atoms contained in the molecule to be predicted and the chemical bonds between the atoms.

The prediction method disclosed in the present invention can determine the specified subgraph through the graph neural network model. In general, the graph neural network model is used to process the graph structure, and the graph structure can accurately characterize the molecular structure of the molecule to be predicted. Based on this, the molecular graph can be determined according to the molecular structure of the molecule to be predicted.

Specifically, for each molecule, the molecule contains atoms, and chemical bonds exist between the atoms, wherein the chemical bonds include ionic bonds and covalent bonds.

The following description is made by taking the execution subject as an example of a server, and the server can determine the molecule to be predicted. The molecule to be predicted can be carried in the prediction request received by the server, or can be carried in the prediction task generated for the molecule to be predicted according to the preset prediction conditions. Then the server can parse the received prediction request or the generated prediction task to determine the molecule to be predicted carried in the prediction request or the prediction task.

Next, the server may determine each atom contained in the molecule to be predicted and the chemical bonds between the atoms according to the molecular structure of the molecule to be predicted.

Finally, the server can determine each node in the molecular graph based on the determined atoms, and then for each chemical bond, determine the edges between the nodes in the molecular graph based on the nodes connected by the chemical bond. In this way, a molecular graph with atoms as nodes and chemical bonds as edges is constructed. The constructed molecular graph contains nodes corresponding to each atom and edges between nodes. Among them, the edges between nodes in the molecular graph are used to characterize the chemical bonds between two nodes connected to the edge in the molecular graph. Take Figure 2 as an example.

FIG2 is a schematic diagram of a molecular graph provided by the present disclosure. The figure takes a methanol molecule as an example. The server can determine the atoms contained in the methanol molecule: C, H, O, and determine the chemical bonds corresponding to each atom. Then, for each atom, the node corresponding to the atom is determined, and for each chemical bond, the edges between the nodes in the molecular graph are determined according to the nodes connected by the chemical bond.

S102: Input the molecular graph into a pre-trained graph neural network model to obtain several specified subgraphs of the molecular graph output by the graph neural network model, wherein the specified subgraphs correspond to substructures contained in the molecular structure of the molecule to be predicted.

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 molecule to be predicted through a graph neural network model, so as to facilitate the subsequent prediction of the specified properties of the molecule to be predicted based on the specified properties of the specified subgraph.

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

Taking the NaOH molecule as an example, the substructure corresponding to the designated subgraph corresponding to the NaOH molecule may be OH. Then, the properties of the NaOH molecule may be predicted by the designated properties of OH. Thus, OH may be used as a substructure for characterizing the molecular properties corresponding to the NaOH molecule. The designated subgraph corresponding to OH may be a subgraph corresponding to the substructure for characterizing the molecular properties corresponding to the NaOH molecule.

Therefore, the server can input the molecular graph determined in the above step S100 into the pre-trained graph neural network model to obtain several specified subgraphs contained in the molecular graph output by the graph neural network model. Among them, for each specified subgraph, the specified subgraph corresponds to a substructure contained in the molecular structure of the molecule to be predicted.

S104: For each designated sub-graph, a designated feature of the designated sub-graph is determined, and a fusion feature is determined according to the designated feature and the representation features corresponding to the pre-stored designated properties.

In one or more embodiments provided in the present application, after determining the designated subgraph, the server can predict the molecular properties corresponding to the molecule to be predicted based on the designated properties of the substructure corresponding to the designated subgraph. In this process, if the designated features corresponding to the designated subgraph are directly classified to determine the designated properties of the substructure corresponding to the designated subgraph, it may be impossible to explain why the substructure has the designated properties. Therefore, the server can fuse the characterization features corresponding to the pre-stored designated properties and the designated features of the designated subgraph to obtain a fusion result, and determine the designated properties of the substructure corresponding to the designated subgraph based on the fusion result.

Specifically, the server pre-stores characterization vectors (i.e., characterization features) corresponding to each specified property. The specified property may be toxic, non-toxic, or soluble in water, slightly soluble in water, or insoluble in water. The type corresponding to the specified property may be set as required, and the present disclosure does not limit this.

At the same time, 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 more similar the substructure corresponding to the specified subgraph is. The higher the probability of having the specified property, the higher the probability of having the specified property. The atoms contained in the specified subgraph and the chemical bonds between the atoms determine the specified feature corresponding to the specified subgraph.

Finally, the server may concatenate the pre-stored representation vectors corresponding to the designated properties and the designated features corresponding to the designated sub-graph, and use the concatenation result as the fusion feature.

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 characterize a specified property of a substructure corresponding to the specified subgraph.

In one or more embodiments provided in the present application, after determining the fusion feature, the server may classify the fusion feature through a classification model to determine the specified property of the substructure corresponding to the specified subgraph.

Specifically, a classification model is pre-set in the server, and the classification model is used to determine the specified properties of the substructure corresponding to the specified subgraph.

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

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

S108: Predicting the molecular properties of the molecule to be predicted according to the target classification results corresponding to the designated subgraphs.

In one or more embodiments provided in the present application, the prediction method needs to predict the molecular property of the molecule to be predicted based on the specified property of the specified subgraph, wherein the molecular property can be the specified property of the molecule to be predicted, or the probability that the molecule to be predicted has the specified property.

Specifically, for each designated subgraph, the server may determine a target classification result corresponding to the designated subgraph.

Then, for each designated property, the server may determine the number of designated subgraphs having the designated property as the designated number, and determine whether the designated number exceeds a preset ratio (eg, one-half) of the number of designated subgraphs included in the molecule to be predicted.

If so, the server may determine that the molecule to be predicted has the specified property.

If not, the server may determine that the molecule to be predicted does not have the specified property.

Furthermore, if a molecule contains a toxic substructure, the molecule is likely to be toxic. If the molecule contains three specified subgraphs, of which only one of the substructures corresponding to the specified subgraph is toxic, then obviously, the molecular properties of the molecule to be predicted may not include the specified property of "toxic". In order to avoid the above situation, the server can determine, for each specified property, whether the specified subgraphs contained in the molecule to be predicted contain a specified subgraph with the specified property.

If so, the server may determine that the molecule to be predicted has the specified property.

If not, the server may determine that the molecule to be predicted does not have the specified property.

Of course, the server may also determine, for each designated property, the probability that the molecule to be predicted has the designated property according to the probability that each designated subgraph included in the molecule to be predicted has the designated property.

Therefore, the server can determine the molecular properties of the molecule to be predicted based on the determined specified properties of the molecule to be predicted, or the probability that the molecule to be predicted has each specified property. Among them, the server can directly use the probability of the molecule to be predicted having each specified property as the molecular property of the molecule to be predicted, or can use the specified property whose probability exceeds a preset specific probability as the molecular property of the molecule to be predicted. How to determine the molecular property of the molecule to be predicted based on the target classification results corresponding to each specified subgraph can be set as needed, and the present disclosure does not limit this.

Based on the prediction method shown in FIG1 , based on the molecular structure of the molecule to be predicted, the graph neural network model is used to determine the designated subgraph corresponding to the substructure of the molecule to be predicted, and then the molecular properties of the molecule to be predicted are predicted based on the designated properties of the designated subgraph. It can be seen that the molecule to be predicted has its corresponding molecular properties because the molecule to be predicted The molecule contains substructures with specified properties. Obviously, the prediction method provides explainability for the predicted molecule to have its corresponding molecular properties, ensuring the credibility of the prediction result.

Furthermore, for molecules, the chemical bonds possessed by the molecules can usually affect the corresponding physical properties of the molecules. The purpose of the prediction method provided by the present application is to determine a designated subgraph of a substructure that can be used to characterize the molecular properties of the molecule to be predicted, and then predict the molecular properties of the molecule to be predicted by the designated properties possessed by the designated subgraph. Based on the same idea, if the chemical bonds that can be used to characterize the molecular properties of the molecule to be predicted are determined, and then the designated subgraph is determined based on the determined chemical bonds, then the molecular properties of the molecule to be predicted can be determined based on the target classification results of the substructure corresponding to the designated subgraph.

Specifically, the server can use the molecular graph determined in step S100 as input into a pre-trained graph neural network model to obtain the confidence corresponding to each chemical bond in the molecule to be predicted 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 molecule to be predicted. For each chemical bond, the confidence corresponding to the chemical bond is used to characterize the probability that the edge corresponding to the chemical bond belongs to the specified subgraph.

Then, the server may determine the designated edges belonging to the designated subgraph according to the determined confidence levels.

Finally, the server may determine the designated subgraph according to the determined designated edges and the nodes connecting the designated edges.

Furthermore, for each chemical bond, the properties of the chemical bond can be characterized based on the properties of the atoms connecting the chemical bond. Therefore, when determining the confidence corresponding to each chemical bond, the server can also determine the bond feature corresponding to each chemical bond, and then determine the confidence corresponding to the chemical bond based on the bond feature.

Specifically, the server may determine the node features corresponding to each node in the molecular graph.

Therefore, for each chemical bond contained in the molecule to be predicted, the server may determine two nodes connected by the edge corresponding to the chemical bond, and perform feature extraction on the two nodes respectively to determine the node features of the two nodes.

Then, the server may concatenate the node features of the two nodes as the bond features of the chemical bond.

Finally, the server can 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 model.

Of course, the server may also perform feature extraction on the chemical bond, determine the initial feature corresponding to the chemical bond, and then fuse the initial feature corresponding to the chemical bond with the node features of the two nodes, and use the fusion result as the bond feature of the chemical bond.

In addition, in the present disclosure, for each atom in the molecular graph, the property corresponding to the atom is not only affected by the atom itself, but also by other atoms connected to the atom. Therefore, when determining the node characteristics of the nodes corresponding to each atom, it can also be determined based on the characteristics of the neighboring nodes of the node.

Specifically, for each node and each edge included in the molecular graph, the server may extract features of the node and the edge to determine the initial features of the node and the initial features of the edge.

Then, for each node in the molecular graph, the server may determine the neighbor nodes of the node.

Finally, the server may determine the node characteristics of the node based on the initial characteristics of the node, the initial characteristics of the neighboring node, and the initial characteristics of the edge between the neighboring node and the node.

Of course, it should be noted that the server can also use the determined node feature of the node as the initial feature of the node, and continue to determine the node feature of the node based on the initial feature of the node and the initial features of the node's neighbor nodes according to the re-determined initial feature of the node. In this way, the purpose of transferring the properties of each atom in the molecular graph along each chemical bond is achieved, and the accuracy of the confidence of the chemical bond determined based on the node feature of the node is further guaranteed.

Furthermore, in the present application, the purpose of determining the fusion feature is to determine the specified property of the specified feature based on the similarity between the specified feature and the characterization features corresponding to each pre-stored specified property. For each specified property, if the specified feature is enhanced based on the similarity between the characterization feature corresponding to the specified property and the specified feature, the accuracy of the target classification result determined based on the enhanced result will be higher. Based on this, the server can enhance the specified feature based on the characterization feature of the specified property.

Specifically, the server may determine, for each specified property, that the specified property corresponds to the specified subgraph according to the similarity between the characterizing feature of the specified property and the specified feature of the specified subgraph, and the characterizing feature of the specified property. Enhanced features of graphs.

After determining that each designated property corresponds to the enhanced feature of the designated sub-graph, the server may fuse the designated feature with the enhanced feature of each designated property corresponding to the designated sub-graph to obtain a fused feature. Subsequently, the target classification result of the designated sub-graph may be obtained based on the fused feature.

Wherein, for each specified property, the characterization feature of the specified property can be determined in the following manner:

For each specified property, the sample molecule graphs corresponding to each sample molecule of the specified property are input into a pre-trained graph neural network model to obtain the specified subgraphs corresponding to each sample molecule of the specified property output by the graph neural network model, and then the characterization features of the specified property are determined based on the specified features corresponding to each specified subgraph.

Of course, the server can also select any sample molecule from the sample molecules corresponding to the specified property, input the molecular graph of the sample molecule into the pre-trained graph neural network model, obtain the specified subgraph corresponding to the sample molecule output by the graph neural network model, and then use the specified features of the specified subgraph as the characterization features of the specified property.

How to determine the characterization features corresponding to each specified property can be set according to needs, and this disclosure does not impose any restrictions on this.

Furthermore, for the molecule to be predicted, in addition to the substructures corresponding to the specified subgraphs, the molecular structure of the molecule to be predicted also includes other substructures, and the properties of the other substructures may also affect the properties of the molecule to be predicted. Therefore, the server may also use the other substructures as specific substructures and determine the properties of the molecule to be predicted based on the specific substructures.

Specifically, the server may determine, according to the molecular graph and the designated graphs, other substructures in the molecular structure of the molecule to be predicted except for the substructures corresponding to the designated subgraphs as specific substructures.

Then, the server may determine the specific subgraph corresponding to the specific substructure, and determine the specific features of the specific subgraph.

Next, the server may input the specific feature into the classification model to obtain a specific classification result output by the classification model.

Finally, the server can predict the molecular properties of the molecule to be predicted based on the specific classification result and the target classification results corresponding to each designated subgraph, as shown in FIG3 .

FIG3 is a flow chart of the prediction method provided by the present disclosure, wherein the server can use the molecular graph of the molecule to be predicted as input, input it into the graph neural network model, and obtain the specified subgraph and the specific subgraph output by the graph neural network model. The server can then fuse the specified features of the specified subgraph with the preset characterization features of each specified property to obtain a fused feature, and then input the fused feature and the specific subgraph into the classification model respectively, to obtain the target classification result corresponding to the fused feature output by the classification model, and the specific classification result corresponding to the specific subgraph. Finally, the server can determine the prediction result based on the specific classification result and the target classification result, and the prediction result is the molecular property of the molecule to be predicted.

In addition, the graph neural network model and the classification model in the present disclosure can be trained in the following manner:

Specifically, the server may obtain a number of sample molecules that have been labeled with specified properties. For each sample molecule that has been labeled with specified properties, according to the atoms contained in the sample molecule and the chemical bonds between the atoms, a sample molecule graph with atoms as nodes and chemical bonds as edges is established as a training sample. At the same time, the server may use the specified properties as the label of the training sample.

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

Next, the server may determine the sample features corresponding to the designated subgraphs of each sample, and determine the characterization features of each designated property according to the sample features and the annotations corresponding to the training samples.

Then, the server may determine the fusion features corresponding to each training sample according to the features of each sample and the characterization features of each specified property.

Afterwards, the server can input each fusion feature into the classification model to be trained to obtain the sample classification result output by the classification model.

Therefore, the server can determine the sample classification results of the sample specified sub-graph corresponding to each training sample. Determine the sample properties corresponding to each training sample.

Finally, the server can determine the loss according to the sample properties and labels corresponding to each training sample, and train the graph neural network model and the classification model with the goal of minimizing the loss.

Furthermore, the graph neural network in the present disclosure can also be trained in the following manner:

Specifically, the server may obtain a number of sample molecules with specified properties labeled, and for each sample molecule with specified properties labeled, according to the atoms contained in the sample molecule and the chemical bonds between the atoms, establish a sample molecule graph with atoms as nodes and chemical bonds as edges as training samples.

Secondly, the server may determine, for each training sample, a designated subgraph included in the training sample as a target sample subgraph of the training sample.

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

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

Based on the same idea, the present disclosure also provides a prediction device, as shown in FIG4 .

FIG4 is a prediction device provided by the present disclosure, wherein:

The first determination module 200 is used to establish a molecular graph with atoms as nodes and chemical bonds as edges according to the obtained atoms contained in the molecule to be predicted and the chemical bonds between the atoms.

The second determination module 202 is used to input the molecular graph into a pre-trained graph neural network model to obtain several specified subgraphs of the molecular graph output by the graph neural network model, wherein the specified subgraphs correspond to substructures contained in the molecular structure of the molecule to be predicted.

The fusion module 204 is used to determine, for each designated sub-graph, a designated feature of the designated sub-graph, and determine a fusion feature according to the designated feature and the pre-stored representation features corresponding to each designated property.

The classification module 206 is used to input the fusion feature into the classification model to obtain the target classification result output by the classification model, and the target classification result is used to characterize the specified property of the substructure corresponding to the specified subgraph.

The prediction module 208 is used to predict the molecular properties of the molecule to be predicted according to the target classification results corresponding to each designated subgraph.

Optionally, the first determination module 200 is used to input the molecular graph into a pre-trained graph neural network model to obtain the confidence corresponding to each chemical bond in the molecule to be predicted output by the graph neural network model, wherein the confidence is used to characterize the probability that the edge corresponding to the chemical bond belongs to a specified subgraph, and based on each confidence, determine each specified edge belonging to the specified subgraph, and determine the specified subgraph based on each specified edge and the nodes connecting the each specified edge.

Optionally, the first determination module 200 is used to determine the node features corresponding to each node in the molecular graph, and for each chemical bond contained in the molecule to be predicted, determine the bond features of the chemical bond according to the node features of the nodes connected by the edges corresponding to the chemical bond, and input the bond features into a pre-trained graph neural network model to obtain the confidence of the chemical bond output by the graph neural network model.

Optionally, the first determination module 200 is used to extract features from each node and each edge contained in the molecular graph, determine the initial features corresponding to each node and the initial features corresponding to each edge, determine the neighbor nodes of each node in the molecular graph, and determine the node features of the node based on 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.

Optionally, the fusion module 204 is used to determine, for each specified property, an enhanced feature of the specified property corresponding to the specified sub-graph based on the similarity between the specified feature and the characterization feature of the specified property, and the characterization feature of the specified property, and fuse the specified feature and the enhanced features of the specified sub-graph corresponding to each specified property to obtain a fused feature.

Optionally, the prediction module 208 is used to determine, based on the molecular graph and each designated subgraph, other substructures in the molecular structure of the molecule to be predicted except for each substructure corresponding to each designated subgraph as specific substructures, and determine the A specific subgraph corresponding to the specific substructure is determined, and specific features of the specific subgraph are determined. The specific features are input into the classification model to obtain a specific classification result output by the classification model, and the molecular properties of the molecule to be predicted are predicted according to the specific classification result and the target classification results corresponding to each designated subgraph.

The device also includes:

The training module 210 is used to train the graph neural network model and the classification model in the following manner: for each sample molecule that has been labeled with a specified property, a sample molecule graph with atoms as nodes and chemical bonds as edges is established as a training sample based on the atoms contained in the sample molecule and the chemical bonds between the atoms, and the specified property is used as the annotation of the training sample. Each training sample is input into the graph neural network model to be trained to obtain a sample specified subgraph corresponding to each training sample output by the graph neural network model, determine the sample features corresponding to each sample specified subgraph, and determine the characterization features of each specified property based on each sample feature and the annotation corresponding to each training sample, determine the fusion features corresponding to each training sample based on each sample feature and the characterization features of each specified property, input each fusion feature into the classification model to be trained to obtain a sample classification result output by the classification model, determine the sample properties corresponding to each training sample based on the sample classification result of the sample specified subgraph corresponding to each training sample, and train the graph neural network model and the classification model based on the sample properties corresponding to each training sample and their annotations.

The present disclosure also provides a computer-readable storage medium, which stores a computer program. The computer program can be used to execute the prediction method provided in FIG. 1 above.

The present disclosure also provides a schematic structural diagram of an electronic device as shown in FIG5. As shown in FIG5, 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 may also include hardware required for other services. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to implement the prediction method described in FIG1 above. Of course, in addition to the software implementation, the present disclosure does not exclude other implementation methods, such as logic devices or a combination of software and hardware, etc., that is to say, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

In the 1990s, it was very clear whether the improvement of a technology was hardware improvement (for example, improvement of the circuit structure of diodes, transistors, switches, etc.) or software improvement (improvement of the method flow). However, with the development of technology, many improvements of the method flow today can be regarded as direct improvements of the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved method flow into the hardware circuit. Therefore, it cannot be said that the improvement of a method flow cannot be implemented with hardware entity modules. For example, a programmable logic device (PLD) (such as a field programmable gate array (FPGA)) is such an integrated circuit whose logical function is determined by the user's programming of the device. Designers can "integrate" a digital system on a PLD by programming themselves, without having to ask chip manufacturers to design and make dedicated integrated circuit chips. Moreover, nowadays, instead of manually making integrated circuit chips, this kind of programming is mostly implemented by "logic compiler" software, which is similar to the software compiler used when developing and writing programs. The original code before compilation must also be written in a specific programming language, which is called Hardware Description Language (HDL). There is not only one kind of HDL, but many kinds, 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 ones are VHDL (Very-High-Speed Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also know that it is only necessary to program the method flow slightly in the above-mentioned hardware description languages and program it into the integrated circuit, and then it is easy to obtain the hardware circuit that realizes the logic method flow.

The controller may be implemented in any suitable manner, for example, the controller may take the form of a microprocessor or processor and a computer-readable medium storing a computer-readable program code (such as software or firmware) executable by the (micro)processor, a logic gate, a switch, an application specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller. Examples of controllers 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 also know that in addition to implementing the controller in a purely computer-readable program code, the controller can be made to implement the same function in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, this controller can be considered as a hardware component, and the devices included therein for implementing various functions can also be regarded as structures within the hardware component. Or even, the devices for implementing various functions can be regarded as both software modules for implementing the method and structures within the hardware component.

The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may 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.

For the convenience of description, the above device is described in terms of functions and is divided into various units and described separately. Of course, when implementing the present disclosure, the functions of each unit can be implemented in the same or multiple software and/or hardware.

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

The present disclosure is described with reference to the flowcharts and/or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present disclosure. It should be understood that each process and/or box in the flowchart and/or block diagram, as well as the combination of the processes and/or boxes in the flowchart and/or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable lesion detection device to produce a machine, so that the instructions executed by the processor of the computer or other programmable lesion detection device produce a device for implementing the functions specified in one process or multiple processes in the flowchart and/or one box or multiple boxes in the block diagram.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable lesion detection device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device that implements the functions specified in one or more processes in the flowchart and/or one or more boxes in the block diagram.

These computer program instructions may also be loaded onto a computer or other programmable lesion detection device so that a series of operating steps are executed on the computer or other programmable device to produce computer-implemented processing, whereby the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more processes in a flowchart and/or one or more boxes in a block diagram.

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

Memory may include non-permanent storage in a computer-readable medium, in the form of random access memory (RAM) and/or non-volatile memory, such as read-only memory (ROM) or flash memory (flash RAM). Memory is an example of a computer-readable medium.

Computer-readable media include permanent and non-permanent, removable and non-removable media that can implement information storage by any method or technology. 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 technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used Storing information that can be accessed by a computing device. As defined herein, computer-readable media does not include transitory media such as modulated data signals and carrier waves.

It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

It will be appreciated by those skilled in the art that the embodiments of the present disclosure may be provided as methods, systems or computer program products. Therefore, the present disclosure may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, the present disclosure may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

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

Each embodiment in the present disclosure is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

The above description is only an embodiment of the present disclosure and is not intended to limit the present disclosure. For those skilled in the art, the present disclosure may have various modifications and variations. Any modification, equivalent substitution, improvement, etc. made within the spirit and principle of the present disclosure shall be included in the scope of the claims of the present disclosure.

Claims (10)

A prediction method, characterized in that the method comprises: According to the obtained atoms contained in the molecule to be predicted and the chemical bonds between the atoms, a molecular graph is established with the atoms as nodes and the chemical bonds as edges; 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, wherein the specified subgraphs correspond to substructures contained in the molecular structure of the molecule to be predicted; For each designated sub-graph, a designated feature of the designated sub-graph is determined, and a fusion feature is determined according to the designated feature and the representation features corresponding to the pre-stored designated properties; Inputting 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 characterize a specified property of a substructure corresponding to the specified subgraph; The molecular properties of the molecules to be predicted are predicted according to the target classification results corresponding to the designated subgraphs. The method according to claim 1, characterized in that the molecular graph is input into a pre-trained graph neural network model to obtain several specified subgraphs of the molecular graph output by the graph neural network model, specifically comprising: Inputting the molecular graph into a pre-trained graph neural network model, obtaining the confidence corresponding to each chemical bond in the molecule to be predicted output by the graph neural network model, wherein the confidence is used to characterize the probability that the edge corresponding to each chemical bond belongs to a specified subgraph; According to the confidences, the designated edges belonging to the designated subgraph are determined, and according to the designated edges and the nodes connected by the designated edges, the designated subgraph is determined. The method of claim 2, characterized in that the molecular graph is input into a pre-trained graph neural network model to obtain the confidence corresponding to each chemical bond in the molecule to be predicted output by the graph neural network model, specifically comprising: Determine the node features corresponding to each node in the molecular graph; For each chemical bond contained in the molecule to be predicted, determining the bond feature of the chemical bond according to the node features of the nodes connected by the edge corresponding to the chemical bond; The bond feature is input into the pre-trained graph neural network model to obtain the confidence of the chemical bond output by the graph neural network model. The method according to claim 3, characterized in that determining the node features corresponding to each node in the molecular graph specifically comprises: Extracting features from each node and each edge in the molecular graph to determine initial features corresponding to each node and initial features corresponding to each edge; For each node in the molecular graph, the neighbor nodes of the node are determined, and the node features of the node are determined 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. The method according to claim 1, characterized in that determining the fusion feature according to the designated feature and the representation features corresponding to the pre-stored designated properties respectively comprises: For each designated property, determining, according to the similarity between the designated feature and the characterizing feature of the designated property, and the characterizing feature of the designated property, an enhanced feature of the designated property corresponding to the designated subgraph; The designated feature and the enhanced features of each designated property corresponding to the designated sub-graph are fused to obtain the fused feature. The method according to claim 1, characterized in that predicting the molecular properties of the molecule to be predicted according to the target classification results corresponding to each designated subgraph respectively comprises: According to the molecular graph and the designated subgraphs, determining other substructures in the molecular structure of the molecule to be predicted except for the substructures corresponding to the designated subgraphs as specific substructures; Determining a specific subgraph corresponding to the specific substructure, and determining specific features of the specific subgraph; Inputting the specific feature into the classification model to obtain a specific classification result output by the classification model; The molecular properties of the molecule to be predicted are predicted according to the specific classification results and the target classification results respectively corresponding to the designated subgraphs. The method according to claim 1, characterized in that the graph neural network model and the classification model are trained in the following manner: For each sample molecule that has been labeled with a specified property, a sample molecule graph is established with each atom as a node and each chemical bond as an edge according to each atom contained in the sample molecule and each chemical bond between the atoms, as a training sample, and the specified property is used as a label for the training sample; Input each training sample into the graph neural network model to be trained, and obtain a number of sample-specified subgraphs corresponding to each training sample output by the graph neural network model to be trained; Determine the sample features corresponding to the designated subgraphs of each sample, and determine the characterization features of each designated property according to the sample features and the labels corresponding to the training samples; Determine, according to the features of each sample and the characterization features of each specified property, the fusion features corresponding to each training sample; Inputting each fusion feature into the classification model to be trained to obtain a sample classification result output by the classification model to be trained; Determine the sample properties corresponding to each training sample according to the sample classification results of the sample designated subgraphs corresponding to each training sample; The graph neural network model and the classification model are trained according to the sample properties and annotations corresponding to each training sample. A prediction device, characterized in that the device comprises: A first determination module is used to establish a molecular graph with each atom as a node and each chemical bond as an edge according to each atom contained in the acquired molecule to be predicted and each chemical bond between the atoms; A second determination module is used 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, wherein the specified subgraphs correspond to substructures contained in the molecular structure of the molecule to be predicted; A fusion module, used for determining, for each specified sub-graph, a specified feature of the specified sub-graph, and determining a fusion feature according to the specified feature and the representation features corresponding to the pre-stored specified properties; A classification module, used for inputting 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 characterize a specified property of a substructure corresponding to the specified subgraph; The prediction module is used to predict the molecular properties of the molecule to be predicted according to the target classification results corresponding to each designated subgraph. A computer-readable storage medium stores a computer program, wherein the computer program, when executed by a processor, implements the method according to any one of claims 1 to 7. An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the method described in any one of claims 1 to 7 when executing the program.