Method and device for determining drug characteristic information based on artificial intelligence

Abstract

The present invention discloses a method and device for determining drug characteristic information based on artificial intelligence, which relates to the field of intelligent medical technology, and the main purpose is to solve the problem of poor accuracy in determining existing drug characteristic information. It mainly includes: obtaining drug molecular structure image data of the target drug; classifying and processing the drug molecular structure image data based on a trained image classification model to obtain a molecular structure image classification result, wherein the image classification model is trained by mining molecular structure node features based on a self-attention mechanism and layering and pooling the graph convolution network through auxiliary information of the graph topology structure; calling a drug characteristic processing flow that matches the molecular structure image classification result; performing a feature matching operation on the molecular structure image classification result based on the drug characteristic processing flow to obtain the drug characteristic information of the target drug. It is mainly used for determining drug characteristic information based on artificial intelligence.

Description

Technical Field

[0001] The present invention relates to the field of intelligent medical technology, and in particular to a method and device for determining drug characteristic information based on artificial intelligence. Background Art

[0002] In recent years, the application field of intelligent medical technology has gradually developed from clinical treatment to drug research and development. More and more artificial intelligence technologies are involved in the analysis of the applicability of drugs to different diseases, so as to accurately find drugs suitable for clinical treatment. In particular, the molecular structure of drugs is studied to determine the treatment plan or disease treatment suitable for patients based on drug characteristics. At present, existing research based on drug molecular structure uses physical experiments to determine drug characteristics. However, such drug molecular structure recognition process is slow and cannot be effectively used in clinical treatment, which makes the use of drug feature matching in intelligent medical care less efficient. Therefore, there is an urgent need for a drug feature information determination method based on artificial intelligence to solve the above problems. Summary of the invention

[0003] In view of this, the present invention provides a method and device for determining drug characteristic information based on artificial intelligence, the main purpose of which is to solve the problem of poor accuracy in determining existing drug characteristic information.

[0004] According to one aspect of the present invention, a method for determining drug characteristic information based on artificial intelligence is provided, comprising:

[0005] Obtain drug molecular structure image data of target drugs;

[0006] Classify the drug molecular structure image data based on a trained image classification model to obtain a molecular structure image classification result, wherein the image classification model is trained by mining molecular structure node features based on a self-attention mechanism and performing hierarchical pooling on a graph convolutional network through auxiliary information of a graph topology structure;

[0007] Retrieving a drug feature processing flow that matches the molecular structure image classification result;

[0008] Based on the drug feature processing flow, a feature matching operation is performed on the molecular structure image classification result to obtain drug feature information of the target drug.

[0009] Furthermore, before the drug molecular structure image data is classified based on the trained image classification model to obtain the molecular structure image classification result, the method further includes:

[0010] Construct a graph convolutional network based on the number of nodes, adjacency matrix, and feature matrix of the drug molecular structure;

[0011] Determine a self-attention mechanism for graph topology enhancement, and introduce the self-attention mechanism into the input layer of the graph convolutional network to mine molecular structure node features;

[0012] Based on the graph topology structure obtained by parsing the drug molecular structure image data, auxiliary information is determined, and each network layer of the graph convolutional network introducing the self-attention mechanism is pooled based on the auxiliary information, wherein the auxiliary information includes global structure information and local structure information of the drug molecular structure image data;

[0013] The graph convolutional network after pooling is trained using molecular structure image sample data to obtain an image classification model.

[0014] Furthermore, the self-attention mechanism for determining graph topology enhancement includes:

[0015] Based on the elementary residual, a jump connection is constructed for the input layer of the graph convolutional network, and the weight matrix is ​​normalized through identity mapping. The normalized graph convolutional network is combined with a self-attention mechanism to obtain a self-attention mechanism with graph topology enhancement.

[0016] Furthermore, before calling the drug feature processing flow that matches the molecular structure image classification result, the method further includes:

[0017] Receiving a molecular classification task triggered by the drug molecular structure image data, wherein the molecular classification task is used to characterize a corresponding processing node in the drug feature processing flow executed on different drug molecular structure image data;

[0018] The process of retrieving drug features matching the molecular structure image classification result includes:

[0019] Parsing the processing node of the drug molecular structure image data in the molecular classification task, and matching the molecular structure image classification result obtained based on the classification processing with the processing node, wherein the processing node is at least one of executing drug molecular composition features, drug molecular attribute features, and drug molecular structure matching in the drug feature processing flow;

[0020] At least one processing node matching the molecular structure image classification result is retrieved.

[0021] Furthermore, the performing of a feature matching operation on the molecular structure image classification result based on the drug feature processing flow to obtain drug feature information of the target drug includes:

[0022] Calculating the similarity between the molecular structure image classification result and the drug molecular composition characteristics, the drug molecular property characteristics and the drug molecular structure;

[0023] If the first similarity between the molecular structure image classification result and the drug molecular composition feature is greater than a preset first similarity threshold, determining the drug feature information including the drug molecular composition feature; and / or,

[0024] If the second similarity between the molecular structure image classification result and the drug molecular attribute feature is greater than a preset second similarity threshold, determining the drug feature information including the drug molecular attribute feature; and / or,

[0025] If the third similarity between the molecular structure image classification result and the drug molecular structure is greater than a preset third similarity threshold, drug feature information including the drug molecular structure is determined.

[0026] Furthermore, after performing a feature matching operation on the molecular structure image classification result based on the drug feature processing flow to obtain drug feature information of the target drug, the method further includes:

[0027] Obtaining a disease characteristic database, and determining whether the drug characteristic information and each disease characteristic information in the disease characteristic database have antagonistic attributes;

[0028] If it has the antagonistic attribute, the disease characteristic information is output.

[0029] Furthermore, the determining whether the drug feature and each disease feature information in the disease feature database have adversarial attributes includes:

[0030] Obtaining a confrontation attribute list, wherein the confrontation attribute list records biological characteristic information and chemical characteristic information of different disease characteristic information, and marks indicating whether there is a medical association between each of the biological characteristic information and different drug effect information;

[0031] Whether the drug characteristic information has antagonistic properties is determined based on the marks corresponding to the biological characteristic information and the chemical characteristic information.

[0032] According to another aspect of the present invention, there is provided a device for determining drug characteristic information based on artificial intelligence, comprising:

[0033] An acquisition module, used to acquire drug molecular structure image data of a target drug;

[0034] A processing module, used to classify the drug molecular structure image data based on a trained image classification model to obtain a molecular structure image classification result, wherein the image classification model is trained by mining molecular structure node features based on a self-attention mechanism and performing hierarchical pooling on a graph convolutional network through auxiliary information of a graph topology structure;

[0035] A retrieval module, used to retrieve a drug feature processing flow that matches the molecular structure image classification result;

[0036] A matching module is used to perform a feature matching operation on the molecular structure image classification result based on the drug feature processing flow to obtain drug feature information of the target drug.

[0037] Furthermore, the device also includes:

[0038] A building module for constructing a graph convolutional network based on the number of nodes, adjacency matrix, and feature matrix of the drug molecular structure;

[0039] A determination module, used to determine a self-attention mechanism for graph topology enhancement, and introduce the self-attention mechanism into the input layer of the graph convolutional network to mine molecular structure node features;

[0040] A parsing module, configured to determine auxiliary information based on the graph topology structure parsed from the drug molecular structure image data, and to pool each network layer of the graph convolutional network that introduces the self-attention mechanism based on the auxiliary information, wherein the auxiliary information includes global structure information and local structure information of the drug molecular structure image data;

[0041] The training module is used to train the graph convolutional network after pooling through molecular structure image sample data to obtain an image classification model.

[0042] Furthermore, the determination module is specifically used to construct a jump connection for the input layer of the graph convolutional network based on elementary residuals, and to normalize the weight matrix through identity mapping, and to combine the normalized graph convolutional network with the self-attention mechanism to obtain a self-attention mechanism with graph topology enhancement.

[0043] Furthermore, the device also includes:

[0044] A receiving module receives a molecular classification task triggered by the drug molecular structure image data, wherein the molecular classification task is used to characterize a corresponding processing node in the drug feature processing flow executed on different drug molecular structure image data;

[0045] The calling module includes:

[0046] A parsing unit, used for parsing the processing node of the drug molecular structure image data in the molecular classification task, and matching the molecular structure image classification result obtained based on the classification processing with the processing node, wherein the processing node is at least one of executing drug molecular composition feature, drug molecular attribute feature and drug molecular structure matching in the drug feature processing flow;

[0047] The retrieving unit is used to retrieve at least one processing node that matches the molecular structure image classification result.

[0048] Furthermore, the matching module includes:

[0049] A calculation unit, used for calculating the similarity between the molecular structure image classification result and the drug molecular composition feature, the drug molecular property feature and the drug molecular structure;

[0050] A first determining unit is configured to determine drug feature information including the drug molecular composition feature if a first similarity between the molecular structure image classification result and the drug molecular composition feature is greater than a preset first similarity threshold; and / or,

[0051] A second determination unit is configured to determine drug feature information including the drug molecular attribute feature if a second similarity between the molecular structure image classification result and the drug molecular attribute feature is greater than a preset second similarity threshold; and / or,

[0052] The third determining unit is used to determine the drug characteristic information including the drug molecular structure if the third similarity between the molecular structure image classification result and the drug molecular structure is greater than a preset third similarity threshold.

[0053] Furthermore, the device also includes:

[0054] A judgment module, used to obtain a disease characteristic database, and judge whether the drug characteristic information and each disease characteristic information in the disease characteristic database have antagonistic attributes;

[0055] An output module is used to output the disease characteristic information if it has the antagonistic attribute.

[0056] Furthermore, the judging module includes:

[0057] An acquisition unit, used for acquiring a confrontation attribute list, wherein the confrontation attribute list records biological characteristic information and chemical characteristic information of different disease characteristic information, and marks indicating whether there is a medical association between each of the biological characteristic information and different drug effect information;

[0058] A judgment unit is used to judge whether the drug characteristic information has antagonistic properties based on the marks corresponding to the biological characteristic information and the chemical characteristic information.

[0059] According to another aspect of the present invention, a storage medium is provided, wherein at least one executable instruction is stored in the storage medium, and the executable instruction enables a processor to perform operations corresponding to the above-mentioned artificial intelligence-based drug characteristic information determination method.

[0060] According to another aspect of the present invention, there is provided a computer device, comprising: a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus;

[0061] The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform operations corresponding to the above-mentioned artificial intelligence-based drug characteristic information determination method.

[0062] By means of the above technical solution, the technical solution provided by the embodiment of the present invention has at least the following advantages:

[0063] The present invention provides a method and device for determining drug characteristic information based on artificial intelligence. Compared with the prior art, the embodiment of the present invention obtains drug molecular structure image data of a target drug; classifies and processes the drug molecular structure image data based on a trained image classification model to obtain a molecular structure image classification result, wherein the image classification model is trained by mining molecular structure node features based on a self-attention mechanism and layering and pooling a graph convolution network through auxiliary information of a graph topology structure; retrieves a drug characteristic processing flow that matches the molecular structure image classification result; performs a feature matching operation on the molecular structure image classification result based on the drug characteristic processing flow to obtain drug characteristic information of the target drug, thereby realizing determination of drug characteristics based on artificial intelligence, greatly accelerating the recognition accuracy of drug characteristics, reducing the complexity and time consumption of manual recognition, and greatly accelerating the speed of determining drug characteristics through recognition of drug molecular structures, thereby improving the effectiveness of matching drug characteristics with symptoms, and realizing an intelligent determination of drug characteristics.

[0064] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented according to the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0066] Figure 1 A flow chart of a method for determining drug characteristic information based on artificial intelligence provided by an embodiment of the present invention is shown;

[0067] Figure 2 A flowchart of another method for determining drug characteristic information based on artificial intelligence provided by an embodiment of the present invention is shown;

[0068] Figure 3 A schematic diagram of a graph convolutional neural network model training structure provided by an embodiment of the present invention is shown;

[0069] Figure 4 A flowchart of another method for determining drug characteristic information based on artificial intelligence provided by an embodiment of the present invention is shown;

[0070] Figure 5 A block diagram of a drug characteristic information determination device based on artificial intelligence provided by an embodiment of the present invention is shown;

[0071] Figure 6 A schematic structural diagram of a computer device provided by an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0072] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0073] The embodiments of the present invention can acquire and process relevant data based on artificial intelligence technology. Artificial Intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0074] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0075] Based on this, in one embodiment, Figure 1 As shown, a method for determining drug characteristic information based on artificial intelligence is provided, and the method is applied to a computer device such as a server as an example for explanation, wherein the server can be an independent server, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms, such as intelligent medical systems, digital medical platforms, etc. The above method includes the following steps:

[0076] 101. Obtain drug molecular structure image data of the target drug.

[0077] In an embodiment of the present invention, the execution subject may be an intelligent management system with an information push function, such as an intelligent medical system, a data medical platform, etc. Exemplarily, the current execution subject is an intelligent medical system, the target drug is a drug suitable for drug characteristics determination, and correspondingly, the drug molecular structure image data of the target drug is a molecule of the target drug represented by a graph structure, wherein the image content in the drug molecular structure image data is the atomic-chemical bond structure of the target drug molecule, and the characteristic content of the molecular structure such as the spatial features, atomic number, charge number, etc. in the form of nodes and edges can be abstracted from the image content, so that a classification implementation method for the drug molecular structure can be obtained based on the classification of the image data, that is, through the graph neural network, the local relationship of the graph can be captured by transmitting specific information such as nodes and edges, and the graph attributes can be automatically learned, so as to efficiently perform the graph classification task.

[0078] It should be noted that the drug molecular structure image data in the embodiment of the present invention is obtained by loading the drug molecular structure image data of the target drug generated by the intelligent medical system as the current execution subject based on the computer software for making molecular structure diagrams. At this time, the operator can obtain the drug molecular structure image data matching the target drug based on the drug database already stored in the current intelligent medical system, or can make it through a molecular structure making application and obtain it in a specified file format in the intelligent medical system. The embodiment of the present invention does not make specific limitations.

[0079] 102. Classify the drug molecular structure image data based on the trained image classification model to obtain a molecular structure image classification result.

[0080] In the embodiment of the present invention, since the drug molecular structure image data contains graph nodes and edges, wherein the graph nodes contain entity information, such as atoms in a compound, and the edges contain relationship information between entities, such as chemical bonds between atoms in the compound image data, in order to classify the drug molecular structure image data and obtain the drug molecular classification results to determine the drug characteristic information, model training is performed in advance to obtain an image classification model to classify the drug molecular structure graph data and obtain the molecular structure image classification results. Among them, since the drug molecular structure image data is classified by a graph neural network, the corresponding molecular structure image classification results are classification results representing different atom-chemical bonds, so as to determine the drug molecular characteristics based on the molecular structure image classification results.

[0081] It should be noted that in order to solve the limitation of insufficient graph pooling, the image classification model is obtained by mining the molecular structure node features based on the self-attention mechanism and performing hierarchical pooling on the graph convolution network through the auxiliary information of the graph topology structure, and completing the training. That is, after constructing the graph convolution neural network, during the training process, based on the adjacency matrix A and the feature matrix X of the graph sample data, a self-attention score is calculated using a Transformer-inspired module as a selection criterion to mine the molecular structure node features. At the same time, the first layer in the graph convolution neural network selects high-scoring nodes with a learnable score function L2Pool to delete unnecessary nodes. At this time, the function L2Pool relies on the auxiliary information of the enhanced graph topology structure in the self-attention and graph convolution network. In this way, while compressing the node scale of the original graph sample data, as much information as possible can be retained, thereby improving the accuracy of the image classification model in graph classification.

[0082] 103. Retrieve a drug feature processing flow that matches the molecular structure image classification result.

[0083] In the embodiment of the present invention, since the classification results of the molecular structure image include the classification results of different atoms-chemical bonds, in order to increase the accuracy of the determination of the drug characteristic information, the classification results of different atoms-chemical bonds correspond to different drug characteristic processing flows. Among them, the drug characteristic processing flow is used to characterize the process of characteristic determination of applicable drugs. The drug characteristic processing flow includes pre-configured drug molecular composition features, drug molecular attribute features, and drug molecular structure processing nodes that match the classification results of different molecular structure images. The processing flow is obtained by arbitrarily combining different processing nodes, thereby starting targeted drug characteristic determination at each processing node, thereby determining whether it has disease resistance, so as to determine it as a therapeutic drug for certain diseases.

[0084] It should be noted that the intelligent medical system in the embodiment of the present invention pre-stores the correspondence between different drug feature processing flows and different molecular structure image classification results, so as to further determine the features of the atom-chemical bond classification results. For example, the classification result of atom a-chemical bond 1 matches the drug feature processing flow of drug molecular composition features and drug molecular attribute features, and then the drug feature processing flow of drug molecular composition features and drug molecular attribute features is retrieved, so as to perform feature matching processing on atom a-chemical bond 1 based on the processing nodes of drug molecular composition features and drug molecular attribute features, and obtain the drug feature information of the target drug for drug molecular composition features and drug molecular attribute features.

[0085] 104. Perform a feature matching operation on the molecular structure image classification result based on the drug feature processing flow to obtain drug feature information of the target drug.

[0086] In an embodiment of the present invention, the feature matching operation is to perform a one-to-one match on the classification results of atoms and chemical bonds contained in the molecular structure image classification results according to the determined drug feature processing flow, that is, to perform similarity calculation based on all existing drug molecular composition features, drug molecular property features, drug molecular structures and atom-chemical bond classification results stored in the intelligent medical system, thereby determining the drug feature information of the target drug.

[0087] In one embodiment of the present invention, in order to further define and illustrate, as Figure 2 As shown, in step 102, before classifying the drug molecular structure image data based on the trained image classification model to obtain the molecular structure image classification result, the method further includes:

[0088] 201. Construct a graph convolutional network based on the number of nodes, adjacency matrix and feature matrix of drug molecular structure;

[0089] 202. Determine a self-attention mechanism for graph topology enhancement, and introduce the self-attention mechanism into an input layer of the graph convolutional network;

[0090] 203. Determine auxiliary information based on the graph topology structure obtained by parsing the drug molecular structure image data, and pool each network layer of the graph convolutional network that introduces the self-attention mechanism based on the auxiliary information;

[0091] 204. The graph convolutional network after pooling is trained using molecular structure image sample data to obtain an image classification model.

[0092] In the embodiment of the present invention, in order to classify the image data, a graph convolutional network is constructed based on the features of the image data. Figure 3 As shown in the figure, a sample data of a molecular structure image to be classified is an image data in the form of graph nodes and edges. Therefore, when constructing a graph convolutional network, it is constructed based on the number of nodes, adjacency matrix, and feature matrix of the drug molecular structure, that is, the image data is represented as G = (A, X), where A and X represent the adjacency matrix and the feature matrix, respectively. The number of graph nodes in the image data is n, and the node feature dimension is d. Then, the graph convolutional network GNN generates an f-dimensional graph convolutional network representation H = [h 1 ,h 2 ,...,h n ] T =GNN(A,X),H∈R n×f At the same time, based on the adjacency matrix A and the feature matrix X, a self-attention score is calculated as a selection criterion through the Transformer-inspired module. In order to ensure that the computational complexity is high, in the embodiment of the present invention, the self-attention mechanism is determined to be four-head attention, that is, MH(Q,K,V)=[O 1 ,...,O h ]W o ; Among them, V, K, Q are fixed single values, and the learning parameter matrices corresponding to Q, K and V are WQ, WK and WV. In addition, dmodel is defined as the output dimension of the multi-head attention function. In order to calculate V through the graph convolutional network (GCN) model, and to avoid the problem of over-smoothing and being limited to the shallow architecture of the graph convolution network GCN, which limits its model performance, V is constructed through GCNII (Graph Convolutional Network via Initial residual and Identity mapping) to explicitly utilize the global structure and capture the information interaction between graph nodes according to the structural dependency of the graph nodes, that is, to complete the mining of molecular structure node features.

[0093] It should be noted that when the self-attention mechanism is introduced into the input layer of the graph convolutional network, that is, the lth layer selects the high-scoring graph node i(l+1)∈R with the learnable score function L2Pool nl+1 i(l+1)∈Rnl+1 to delete unnecessary graph nodes, expressed as: y (l) =L2Pool(Att,H (l) ,A (l) );i (l+1) =top k (y (l) ), where the function L2Pool relies on multi-head attention and GCNII enhanced topological information, top k The () function samples the first k graph nodes by discarding the nodes with lower scores, so as to retain as much information as possible while compressing the scale of the graph nodes of the image data, so as to introduce the pooling of each network layer of the graph convolution network with the self-attention mechanism. Among them, in order to construct multi-scale image data, in each pooling layer, the scale of the graph nodes is reduced by sampling to obtain a coarsened image data. At this time, the importance score is calculated for the graph node based on the sampling to retain the most important first k graph nodes and the connection relationship between them to generate the coarsened image data, so as to realize the pooling of each network layer of the graph convolution network. In this process, the pooling model for pooling each layer obtains a more effective graph representation through the global structure information and local structure information of the drug molecular structure image data, and the auxiliary information includes the global structure information and local structure information of the drug molecular structure image data, that is, the network layers of the graph convolution network with the self-attention mechanism are pooled through the auxiliary information. Finally, the graph convolution network after pooling is trained by the molecular structure image sample data as the model training sample to obtain the image classification model.

[0094] In one embodiment of the present invention, for further limitation and explanation, step 202 determines the self-attention mechanism for graph topology enhancement, including: constructing a jump connection for the input layer of the graph convolutional network based on elementary residuals, and normalizing the weight matrix through an identity mapping, combining the normalized graph convolutional network with the self-attention mechanism to obtain a self-attention mechanism for graph topology enhancement.

[0095] In order to avoid the problem of limiting the model performance due to the shallow architecture of the graph convolutional network due to degree smoothing, the initial residual and the GCNII of the identity mapping are used to construct the V in the four-head attention. Specifically, since GCNII is a GCN with an initial residual connection and an identity mapping, in each layer of the graph convolutional network, the initial residual constructs a skip connection in the input layer, and the weight matrix is ​​normalized by the identity mapping, that is, the identity mapping adds the unit matrix to the weight matrix to combine the normalized graph convolutional network with the self-attention mechanism to increase the network depth of GCNII, prevent over-smoothing and continuously improve the performance of GCNII.

[0096] The self-attention mechanism enhanced by graph topology is defined as: GCHII(H,A)=σ(((1-α)AH+αH 0 )((1-β)I n )+βW)); A 4-layer GCNII model is used to construct the value V, thereby introducing graph topology information to optimize the scores of important graph nodes.

[0097] It should be noted that the input of the graph convolutional neural network is an image data structure with graph nodes or edges, that is, it includes the adjacency matrix A of the image data and the corresponding feature attribute information X. The graph convolutional neural network trains the implicit vector representation of each graph node in the image data based on the graph structure and the input node attributes. The goal is to make the vector representation contain sufficiently powerful expression information so that it can help each graph node to extract information. Finally, the information vector representation of the entire graph can be obtained. For example, for a molecular graph composed of atoms and chemical bonds, the molecular-level information representation of the entire molecular compound is extracted through the characteristics of the atomic nodes and the chemical bond information of the edges between atoms. The main process of graph convolutional neural network model learning is to aggregate and update the neighbor information of the graph nodes in the graph data through iteration. In each iteration, each graph node updates its own information by aggregating the features of neighbor nodes and its own features in the previous layer, and usually performs nonlinear transformation on the aggregated information. By stacking multiple layers of networks, each graph node can obtain neighbor node information within the corresponding number of hops, using a new graph coarsening method based on the Transformer self-attention mechanism and network topology information of image data, such as a coarsening pooling method including node features and graph topology features, which is not specifically limited in the embodiments of the present invention.

[0098] In one embodiment of the present invention, for further definition and explanation, before step 103 calls the drug feature processing flow that matches the molecular structure image classification result, the method further includes: receiving a molecular classification task triggered by the drug molecular structure image data.

[0099] In the embodiment of the present invention, since the drug feature processing flow contains pre-configured drug molecular composition features, drug molecular attribute features, and at least one processing node in the drug molecular structure for matching different molecular structure image classification results, in order to meet the operational requirements of matching different drug feature information, different operators can automatically and flexibly execute the drug feature processing flow based on the intelligent medical system, and pre-configure the molecular classification task for the drug feature processing flow so that the operator can trigger the molecular classification task to execute the triggering of the drug feature processing flow for different drug molecular structures. Among them, the molecular classification task is used to characterize the corresponding processing node in the drug feature processing flow executed on different drug molecular structure image data, that is, different processing nodes can be triggered by different molecular classification tasks, so as to complete the execution of the drug feature processing flow for different molecular structure image classification results in a targeted manner.

[0100] Correspondingly, step 103 calls the drug feature processing flow that matches the molecular structure image classification result, including: parsing the processing node of the drug molecular structure image data in the molecular classification task, and matching the molecular structure image classification result obtained based on the classification processing with the processing node; calling at least one processing node that matches the molecular structure image classification result.

[0101] In an embodiment of the present invention, the processing node is at least one of the drug molecular composition features, drug molecular attribute features and drug molecular structure matching executed in the drug feature processing flow, that is, the processing node includes a drug molecular composition feature processing node, a drug molecular attribute feature processing node, and a drug molecular structure processing node. At this time, since the molecular classification task carries the processing nodes expected to be performed on the drug molecular structure image data before image classification processing, but not all drug molecular structure image data can still be executed according to the processing nodes after image classification, therefore, the processing nodes of the drug molecular structure image data in the molecular classification task are first parsed, and then the molecular structure image classification results obtained based on the classification processing are matched with the processing nodes, so as to retrieve at least one processing node that is finally matched with the molecular structure image classification results. For example, if the processing nodes of the drug molecular structure image data in the molecular classification task of the target drug a are the drug molecular composition feature processing node, the drug molecular attribute feature processing node, and the drug molecular structure processing node, after the drug molecular structure image data is classified and processed, the processing node corresponding to the atom a-chemical bond 1 in the molecular structure image classification result matches the drug molecular composition feature processing node and the drug molecular structure processing node, then the drug molecular composition feature processing node and the drug molecular structure processing node are retrieved to execute the processing node. Among them, for the matching between each molecular structure image classification result and the processing node, a matching relationship list can be pre-entered to accurately match different molecular structure image classification results with the processing nodes, and the embodiment of the present invention does not specifically limit the matching relationship list.

[0102] In one embodiment of the present invention, in order to further define and illustrate, as Figure 4 As shown, step 104 performs a feature matching operation on the molecular structure image classification result based on the drug feature processing flow to obtain drug feature information of the target drug, including:

[0103] 1041. Calculate the similarity between the molecular structure image classification result and the drug molecular composition feature, the drug molecular property feature, and the drug molecular structure;

[0104] 1042. If the first similarity between the molecular structure image classification result and the drug molecular composition feature is greater than a preset first similarity threshold, determine the drug feature information including the drug molecular composition feature; and / or,

[0105] 1043. If the second similarity between the molecular structure image classification result and the drug molecular attribute feature is greater than a preset second similarity threshold, determine the drug feature information including the drug molecular attribute feature; and / or,

[0106] 1044. If the third similarity between the molecular structure image classification result and the drug molecular structure is greater than a preset third similarity threshold, determine drug feature information including the drug molecular structure.

[0107] In order to perform feature matching operations on the molecular structure image classification results and each processing node in the drug feature processing flow, after calling the corresponding processing nodes, such as the drug molecular composition feature processing node, the drug molecular attribute feature processing node, and the drug molecular structure processing node, the corresponding matching operation is performed. In an embodiment of the present invention, the matching operation for each processing node is to perform similarity calculation on the molecular structure image classification results according to each processing node, so as to determine the drug feature information of the target drug based on the obtained similarity. In an embodiment of the present invention, in the drug molecular composition feature processing node, the molecular structure image classification results are calculated for similarity with the drug molecular composition features stored in the intelligent medical system. Since the molecular structure image classification results contain image data of atoms-chemical bonds, the similarity is calculated specifically by performing similarity calculation on the image corresponding to the drug molecular composition feature, so as to determine whether the calculated similarity value is greater than the first similarity threshold. If so, it is determined that the target drug contains drug feature information of the drug molecular composition feature. In the drug molecular attribute feature processing node, by calculating the similarity between the molecular structure image classification result and the drug molecular attribute feature stored in the intelligent medical system, since the molecular structure image classification result contains the image data of the atom-chemical bond, specifically by calculating the similarity with the image corresponding to the drug molecular attribute feature, it is determined whether the calculated similarity value is greater than the second similarity threshold value, and if so, it is determined that the target drug contains the drug feature information of the drug molecular attribute feature. The drug molecular structure processing node calculates the similarity between the molecular structure image classification result and the drug molecular structure stored in the intelligent medical system, since the molecular structure image classification result contains the image data of the atom-chemical bond, specifically by calculating the similarity with the image corresponding to the drug molecular structure, it is determined whether the calculated similarity value is greater than the third similarity threshold value, and if so, it is determined that the target drug contains the drug feature information of the drug molecular structure.

[0108] In addition, the drug molecular composition feature is used to characterize the feature content composed of different molecules in the drug, for example, the molecular composition features including cyclic aromatic hydrocarbons, methyl groups, etc. The drug molecular attribute feature is used to characterize the physical or chemical properties produced by different molecules in the drug, for example, chemical properties such as toluene-easy oxidation, and the drug molecular structure is used to characterize the chemical bond structure between molecules, for example, aromatic hydrocarbons have the basic molecular structure of a benzene ring. Therefore, the molecular structure image classification results can be matched through the processing nodes corresponding to the drug molecular composition features, drug molecular attribute features, and drug molecular structures. In the embodiment of the present invention, in order to improve the matching accuracy, the drug molecular composition features, drug molecular attribute features, and drug molecular structures can also be digitized, and the molecular structure image classification results can be digitized at the same time, so that the drug molecular composition features, drug molecular attribute features, and drug molecular structures to be calculated for similarity are put into a data unit with the molecular structure image classification results for similarity calculation to complete the matching process.

[0109] In one embodiment of the present invention, in order to further limit and illustrate, the step is to perform a feature matching operation on the molecular structure image classification result based on the drug feature processing flow to obtain the drug feature information of the target drug. The method also includes: obtaining a disease feature database, determining whether the drug feature information and each disease feature information in the disease feature database have adversarial attributes; if they have the adversarial attributes, outputting the disease feature information.

[0110] In an embodiment of the present invention, in order to meet the use requirements of the drug characteristic information of the target drug by the intelligent medical system, after obtaining the drug characteristic information, the drug characteristic information and each disease characteristic information can be judged to see whether there is antagonism, so as to determine whether to output the disease characteristic information related to the drug characteristic information to the user, so as to perform the treatment operation of the related disease. Among them, the disease characteristic database stored in the intelligent medical system records the disease characteristic information corresponding to various known diseases. At this time, the disease characteristic information is used to characterize the symptoms caused by the disease to the human body. For example, a certain disease will cause blood pressure higher than 180 mmHg, and a certain disease will cause a significant decrease in adrenaline. At this time, the antagonistic attribute is whether there are attributes that play opposite roles between the drug characteristic information and the disease characteristic information. For example, whether the drug characteristic information has a blood pressure-raising effect on the low blood pressure in the disease characteristic information. If there is an antagonistic attribute, it means that this target drug can be used as a treatment for the disease corresponding to this disease characteristic information or relieve symptoms, and has been output to the user for viewing.

[0111] In one embodiment of the present invention, in order to further limit and illustrate, the step of determining whether the drug characteristic and each disease characteristic information in the disease characteristic database have adversarial attributes includes: obtaining a list of adversarial attributes; and determining whether the drug characteristic information has adversarial attributes based on the marks corresponding to the biological characteristic information and the chemical characteristic information.

[0112] When judging whether the drug characteristic information and the disease characteristic information have antagonistic attributes, the antagonistic attribute list in the intelligent medical system is specifically used. The antagonistic attribute list records the biological characteristic information and chemical characteristic information of different disease characteristic information, and the marks of whether there is medical association between different drug special effect information. This mark is the antagonistic attribute with medical association determined by medical experiments. For example, if there is a mark between the drug molecular attribute feature a and the biological characteristic information of increasing the speed of cell receiving oxygen atoms, it means that the target drug containing the drug characteristic information of the drug molecular attribute feature a can be used to treat diseases with slow cell receiving oxygen atoms. Therefore, it is judged whether there is an antagonistic attribute by the marks corresponding to the drug characteristic information and the biological characteristic information and the chemical characteristic information. Among them, the biological characteristic information and chemical characteristic information of different disease characteristic information are respectively used to be pre-entered into the intelligent medical system. As the characteristic content of each disease, the biological characteristic information is the physiological or biological symptom characteristics displayed by the disease, such as pain, fever, blood value, cell volume, etc., and the chemical characteristic information is the symptom characteristics of the chemical composition displayed by the disease, hormone value, etc., which are not specifically limited in the embodiment of the present invention.

[0113] The embodiment of the present invention provides a method for determining drug characteristic information based on artificial intelligence. Compared with the prior art, the embodiment of the present invention obtains drug molecular structure image data of a target drug; classifies and processes the drug molecular structure image data based on a trained image classification model to obtain a molecular structure image classification result, wherein the image classification model is trained by mining molecular structure node features based on a self-attention mechanism and layering and pooling a graph convolution network through auxiliary information of a graph topology structure; retrieves a drug characteristic processing flow that matches the molecular structure image classification result; performs a feature matching operation on the molecular structure image classification result based on the drug characteristic processing flow to obtain drug characteristic information of the target drug, thereby realizing determination of drug characteristics based on artificial intelligence, greatly accelerating the recognition accuracy of drug characteristics, reducing the complexity and time consumption of manual recognition, and greatly accelerating the speed of determining drug characteristics through recognition of drug molecular structures, thereby improving the effectiveness of matching drug characteristics with symptoms, and realizing an intelligent determination of drug characteristics.

[0114] Furthermore, as a response to the above Figure 1 The embodiment of the present invention provides a drug characteristic information determination device based on artificial intelligence, such as Figure 5 As shown, the device comprises:

[0115] An acquisition module 31 is used to acquire drug molecular structure image data of a target drug;

[0116] A processing module 32 is used to classify the drug molecular structure image data based on a trained image classification model to obtain a molecular structure image classification result, wherein the image classification model is trained by mining molecular structure node features based on a self-attention mechanism and performing hierarchical pooling on a graph convolutional network through auxiliary information of a graph topology structure;

[0117] A retrieval module 33, used to retrieve a drug feature processing flow that matches the molecular structure image classification result;

[0118] The matching module 34 is used to perform a feature matching operation on the molecular structure image classification result based on the drug feature processing flow to obtain drug feature information of the target drug.

[0119] Furthermore, the device also includes:

[0120] A building module for constructing a graph convolutional network based on the number of nodes, adjacency matrix, and feature matrix of the drug molecular structure;

[0121] A determination module, used to determine a self-attention mechanism for graph topology enhancement, and introduce the self-attention mechanism into the input layer of the graph convolutional network to mine molecular structure node features;

[0122] A parsing module, configured to determine auxiliary information based on the graph topology structure parsed from the drug molecular structure image data, and to pool each network layer of the graph convolutional network that introduces the self-attention mechanism based on the auxiliary information, wherein the auxiliary information includes global structure information and local structure information of the drug molecular structure image data;

[0123] The training module is used to train the graph convolutional network after pooling through molecular structure image sample data to obtain an image classification model.

[0124] Furthermore, the determination module is specifically used to construct a jump connection for the input layer of the graph convolutional network based on elementary residuals, and to normalize the weight matrix through identity mapping, and to combine the normalized graph convolutional network with the self-attention mechanism to obtain a self-attention mechanism with graph topology enhancement.

[0125] Furthermore, the device also includes:

[0126] A receiving module receives a molecular classification task triggered by the drug molecular structure image data, wherein the molecular classification task is used to characterize a corresponding processing node in the drug feature processing flow executed on different drug molecular structure image data;

[0127] The calling module includes:

[0128] A parsing unit, used for parsing the processing node of the drug molecular structure image data in the molecular classification task, and matching the molecular structure image classification result obtained based on the classification processing with the processing node, wherein the processing node is at least one of executing drug molecular composition feature, drug molecular attribute feature and drug molecular structure matching in the drug feature processing flow;

[0129] The retrieving unit is used to retrieve at least one processing node that matches the molecular structure image classification result.

[0130] Furthermore, the matching module includes:

[0131] A calculation unit, used for calculating the similarity between the molecular structure image classification result and the drug molecular composition feature, the drug molecular property feature and the drug molecular structure;

[0132] A first determining unit is configured to determine drug feature information including the drug molecular composition feature if a first similarity between the molecular structure image classification result and the drug molecular composition feature is greater than a preset first similarity threshold; and / or,

[0133] A second determination unit is configured to determine drug feature information including the drug molecular attribute feature if a second similarity between the molecular structure image classification result and the drug molecular attribute feature is greater than a preset second similarity threshold; and / or,

[0134] The third determining unit is used to determine the drug characteristic information including the drug molecular structure if the third similarity between the molecular structure image classification result and the drug molecular structure is greater than a preset third similarity threshold.

[0135] Furthermore, the device also includes:

[0136] A judgment module, used to obtain a disease characteristic database, and judge whether the drug characteristic information and each disease characteristic information in the disease characteristic database have antagonistic attributes;

[0137] An output module is used to output the disease characteristic information if it has the antagonistic attribute.

[0138] Furthermore, the judging module includes:

[0139] An acquisition unit, used for acquiring a confrontation attribute list, wherein the confrontation attribute list records biological characteristic information and chemical characteristic information of different disease characteristic information, and marks indicating whether there is a medical association between each of the biological characteristic information and different drug effect information;

[0140] A judgment unit is used to judge whether the drug characteristic information has antagonistic properties based on the marks corresponding to the biological characteristic information and the chemical characteristic information.

[0141] The embodiment of the present invention provides a drug characteristic information determination device based on artificial intelligence. Compared with the prior art, the embodiment of the present invention obtains drug molecular structure image data of a target drug; classifies and processes the drug molecular structure image data based on a trained image classification model to obtain a molecular structure image classification result, wherein the image classification model is trained by mining molecular structure node features based on a self-attention mechanism and layering and pooling a graph convolution network through auxiliary information of a graph topology structure; retrieves a drug characteristic processing flow that matches the molecular structure image classification result; performs a feature matching operation on the molecular structure image classification result based on the drug characteristic processing flow to obtain drug characteristic information of the target drug, thereby realizing the determination of drug characteristics based on artificial intelligence, greatly accelerating the recognition accuracy of drug characteristics, reducing the complexity and time consumption of manual recognition, and greatly accelerating the speed of determining drug characteristics through the recognition of drug molecular structures, thereby improving the effectiveness of matching drug characteristics with symptoms, and realizing an intelligent determination of drug characteristics.

[0142] According to one embodiment of the present invention, a storage medium is provided, wherein the storage medium stores at least one executable instruction, and the computer executable instruction can execute the artificial intelligence-based drug characteristic information determination method in any of the above method embodiments.

[0143] Figure 6 A schematic diagram of the structure of a computer device provided according to an embodiment of the present invention is shown. The specific embodiment of the present invention does not limit the specific implementation of the computer device.

[0144] like Figure 6 As shown, the computer device may include: a processor (processor) 402 , a communication interface (Communications Interface) 404 , a memory (memory) 406 , and a communication bus 408 .

[0145] The processor 402 , the communication interface 404 , and the memory 406 communicate with each other via a communication bus 408 .

[0146] The communication interface 404 is used to communicate with other devices such as clients or other servers.

[0147] Processor 402 is used to execute program 410, and specifically can execute the relevant steps in the above-mentioned embodiment of the drug characteristic information determination method based on artificial intelligence.

[0148] Specifically, the program 410 may include program codes, which include computer operation instructions.

[0149] The processor 402 may be a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present invention. The one or more processors included in the computer device may be processors of the same type, such as one or more CPUs; or processors of different types, such as one or more CPUs and one or more ASICs.

[0150] The memory 406 is used to store the program 410. The memory 406 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0151] The program 410 may be specifically configured to enable the processor 402 to perform the following operations:

[0152] Obtain drug molecular structure image data of target drugs;

[0153] Classify the drug molecular structure image data based on a trained image classification model to obtain a molecular structure image classification result, wherein the image classification model is trained by mining molecular structure node features based on a self-attention mechanism and performing hierarchical pooling on a graph convolutional network through auxiliary information of a graph topology structure;

[0154] Retrieving a drug feature processing flow that matches the molecular structure image classification result;

[0155] Based on the drug feature processing flow, a feature matching operation is performed on the molecular structure image classification result to obtain drug feature information of the target drug.

[0156] Obviously, those skilled in the art should understand that the above modules or steps of the present invention can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, and optionally, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order than here, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0157] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for determining drug characteristic information based on artificial intelligence, It is characterized in that include: Obtain drug molecular structure image data of target drugs; Classify the drug molecular structure image data based on a trained image classification model to obtain a molecular structure image classification result, wherein the image classification model is trained by mining molecular structure node features based on a self-attention mechanism and performing hierarchical pooling on a graph convolutional network through auxiliary information of a graph topology structure; Retrieving a drug feature processing flow that matches the molecular structure image classification result; Based on the drug feature processing flow, a feature matching operation is performed on the molecular structure image classification result to obtain drug feature information of the target drug; Before calling the drug feature processing flow that matches the molecular structure image classification result, the method further includes: Receiving a molecular classification task triggered by the drug molecular structure image data, wherein the molecular classification task is used to characterize a corresponding processing node in the drug feature processing flow executed on different drug molecular structure image data; The process of retrieving drug features matching the molecular structure image classification result includes: Parsing the processing node of the drug molecular structure image data in the molecular classification task, and matching the molecular structure image classification result obtained based on the classification processing with the processing node, wherein the processing node is at least one of executing drug molecular composition features, drug molecular attribute features, and drug molecular structure matching in the drug feature processing flow; At least one processing node matching the molecular structure image classification result is retrieved.

2. The method according to claim 1, It is characterized in that Before classifying the drug molecular structure image data based on the trained image classification model to obtain the molecular structure image classification result, the method further includes: Construct a graph convolutional network based on the number of nodes, adjacency matrix, and feature matrix of the drug molecular structure; Determine a self-attention mechanism for graph topology enhancement, and introduce the self-attention mechanism into the input layer of the graph convolutional network to mine molecular structure node features; Based on the graph topology structure obtained by parsing the drug molecular structure image data, auxiliary information is determined, and each network layer of the graph convolutional network introducing the self-attention mechanism is pooled based on the auxiliary information, wherein the auxiliary information includes global structure information and local structure information of the drug molecular structure image data; The graph convolutional network after pooling is trained using molecular structure image sample data to obtain an image classification model.

3. The method according to claim 2, It is characterized in that The self-attention mechanism for determining graph topology enhancement includes: Based on the elementary residual, a jump connection is constructed for the input layer of the graph convolutional network, and the weight matrix is ​​normalized through identity mapping. The normalized graph convolutional network is combined with a self-attention mechanism to obtain a self-attention mechanism with graph topology enhancement.

4. The method according to claim 1, It is characterized in that The performing of a feature matching operation on the molecular structure image classification result based on the drug feature processing flow to obtain drug feature information of the target drug includes: Calculating the similarity between the molecular structure image classification result and the drug molecular composition characteristics, the drug molecular property characteristics and the drug molecular structure; If the first similarity between the molecular structure image classification result and the drug molecular composition feature is greater than a preset first similarity threshold, determining the drug feature information including the drug molecular composition feature; and / or, If the second similarity between the molecular structure image classification result and the drug molecular attribute feature is greater than a preset second similarity threshold, determining the drug feature information including the drug molecular attribute feature; and / or, If the third similarity between the molecular structure image classification result and the drug molecular structure is greater than a preset third similarity threshold, drug feature information including the drug molecular structure is determined.

5. The method according to claim 1, It is characterized in that After performing a feature matching operation on the molecular structure image classification result based on the drug feature processing flow to obtain drug feature information of the target drug, the method further includes: Obtaining a disease characteristic database, and determining whether the drug characteristic information and each disease characteristic information in the disease characteristic database have antagonistic attributes; If it has the antagonistic attribute, the disease characteristic information is output.

6. The method according to claim 5, It is characterized in that The determining whether the drug feature and each disease feature information in the disease feature database have a confrontational attribute includes: Obtaining a confrontation attribute list, wherein the confrontation attribute list records biological characteristic information and chemical characteristic information of different disease characteristic information, and marks indicating whether there is a medical association between each of the biological characteristic information and different drug effect information; Whether the drug characteristic information has antagonistic properties is determined based on the marks corresponding to the biological characteristic information and the chemical characteristic information.

7. Drug characteristic information determination device based on artificial intelligence, It is characterized in that include: An acquisition module, used to acquire drug molecular structure image data of a target drug; A processing module, used to classify the drug molecular structure image data based on a trained image classification model to obtain a molecular structure image classification result, wherein the image classification model is trained by mining molecular structure node features based on a self-attention mechanism and performing hierarchical pooling on a graph convolutional network through auxiliary information of a graph topology structure; A retrieval module, used to retrieve a drug feature processing flow that matches the molecular structure image classification result; A matching module, used to perform a feature matching operation on the molecular structure image classification result based on the drug feature processing flow to obtain drug feature information of the target drug; The device also includes: A receiving module receives a molecular classification task triggered by the drug molecular structure image data, wherein the molecular classification task is used to characterize a corresponding processing node in the drug feature processing flow executed on different drug molecular structure image data; The calling module includes: A parsing unit, used for parsing the processing node of the drug molecular structure image data in the molecular classification task, and matching the molecular structure image classification result obtained based on the classification processing with the processing node, wherein the processing node is at least one of executing drug molecular composition feature, drug molecular attribute feature and drug molecular structure matching in the drug feature processing flow; The retrieving unit is used to retrieve at least one processing node that matches the molecular structure image classification result.

8. A storage medium storing at least one executable instruction, wherein the executable instruction enables a processor to perform operations corresponding to the artificial intelligence-based drug characteristic information determination method as described in any one of claims 1 to 6.

9. A computer device, include: A processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus; The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform operations corresponding to the artificial intelligence-based drug characteristic information determination method as described in any one of claims 1-6.

Images