Drug resistance prediction method, device, electronic device and storage medium

Abstract

The present application provides a drug resistance prediction method, device, electronic device and storage medium, the drug resistance prediction method comprising: inputting the chemical molecular structure of the drug into the molecular formula encoder of the drug resistance prediction model, performing molecular structure modeling processing on the chemical molecular structure, and outputting the structural feature information of the drug; inputting the whole genome mutation information of the isolate into the hierarchical encoder of the drug resistance prediction model, performing encoding processing and attention mechanism processing on the whole genome mutation information, and outputting the encoding features of the whole genome mutation information; inputting the structural feature information and the encoding features into the classification network layer of the drug resistance prediction model for feature fusion and drug resistance classification processing, and outputting the drug resistance result of the isolate for the drug. In this way, more complex drug resistance patterns can be captured and the drug resistance of the drug can be accurately determined.

Description

Technical Field

[0001] The present application relates to the technical field of drug resistance prediction, and in particular to a drug resistance prediction method, device, electronic device and storage medium. Background Art

[0002] While current drug resistance prediction methods have made some progress in improving the classification performance of specific drugs, they still face significant challenges in generalizing to new or low-resource drugs that lack labeling. For example, when predicting whether MTB is resistant to anti-tuberculosis drugs, existing methods face a problem: if a new or rare drug lacks sufficient experimental data (i.e., labeled samples), these methods' performance is significantly reduced. This is because they rely on large amounts of known data to train the models. However, in the early stages of new drug development, such data is often scarce or even nonexistent, which limits the identification of potential drug resistance risks and leads to technical issues such as inaccurate determination of drug resistance. Summary of the Invention

[0003] In view of this, the purpose of this application is to provide a drug resistance prediction method, device, electronic device and storage medium. By introducing the chemical molecular structure of the drug and learning the structural characteristic information of the drug, the structural characteristic information is jointly modeled with the whole genome mutation representation to capture more complex resistance patterns and accurately determine the drug resistance.

[0004] The present invention provides a method for predicting drug resistance, which includes:

[0005] Inputting the chemical molecular structure of the drug into the molecular formula encoder of the drug resistance prediction model, performing molecular structure modeling processing on the chemical molecular structure, and outputting structural characteristic information of the drug;

[0006] Inputting the whole genome mutation information of the isolate into the hierarchical encoder of the drug resistance prediction model, performing encoding processing and attention mechanism processing on the whole genome mutation information, and outputting the encoding features of the whole genome mutation information;

[0007] The structural feature information and the coding features are input into the classification network layer of the drug resistance prediction model for feature fusion and drug resistance classification processing, and the drug resistance result of the isolate to the drug is output; wherein, the drug resistance prediction model is obtained by iterative training of a deep learning model.

[0008] In a possible implementation manner, performing molecular structure modeling on the chemical molecular structure and outputting structural feature information of the drug includes:

[0009] Encoding the chemical molecular structure to generate initial atomic features of the chemical molecular structure;

[0010] Parsing the chemical molecular structure into a molecular graph structure; wherein the nodes in the molecular graph structure are atoms of the drug and the edges are chemical bonds of the drug;

[0011] Feature modeling is performed based on the nodes, edges and initial atomic features in the molecular graph structure to determine the structural feature information; wherein the structural feature information is a graph representation that integrates the chemical molecular structure, atomic properties and initial atomic features.

[0012] In a possible implementation, the molecular graph structure is:

[0013]

[0014] Among them, V is the set of nodes, E is the set of edges consisting of chemical bonds between two atoms, is the molecular graph structure, For drugs The chemical molecular structure of sequence, For analysis tools.

[0015] In a possible implementation manner, performing feature modeling based on the nodes, edges, and initial features of atoms in the molecular graph structure to determine the structural feature information includes:

[0016] The basic chemical properties of atoms are used as the initial feature vectors of the corresponding nodes, and the chemical bond type corresponding to each edge is used as the feature of the edge;

[0017] Performing feature fusion on the initial feature vector of each node and the initial feature of the atom to determine the target feature vector of each node;

[0018] For each node, the features of the edges between the node and its neighboring nodes and the target feature vectors of the neighboring nodes are aggregated, and the target feature vector of the node is updated based on the aggregated features to determine the updated feature vector of the node;

[0019] The updated feature vector of each node is processed based on a global average pooling function to determine the structural feature information.

[0020] In a possible implementation, the inputting of the structural feature information and the coding feature into the classification network layer of the drug resistance prediction model for information fusion and drug resistance classification processing, and outputting the drug resistance result of the isolate to the drug, includes:

[0021] Performing feature splicing on the structural feature information and the coding feature to determine the spliced ​​feature;

[0022] Determine the fusion features based on the spliced ​​features and attention weights;

[0023] The fusion features are processed based on the feedforward neural sub-network layer and the activation function in the classification network layer, and the drug resistance result of the isolated strain to the drug is output.

[0024] In a possible implementation, the inputting of the structural feature information and the coding feature into the classification network layer of the drug resistance prediction model for feature fusion and drug resistance classification processing, and outputting the drug resistance result of the isolate to the drug, includes:

[0025] Performing feature splicing on the structural feature information and the coding feature to determine the spliced ​​feature;

[0026] Determine the fusion features based on the spliced ​​features and attention weights;

[0027] The fusion features are processed based on the feedforward neural sub-network layer and the activation function in the classification network layer, and the drug resistance result of the isolated strain to the drug is output.

[0028] In one possible implementation, the drug resistance prediction model is determined by the following steps:

[0029] Inputting sample chemical molecular structures of multiple sample drugs and sample whole genome mutation information of sample isolates into the deep learning model, and outputting predicted drug resistance results of the sample isolates for the sample drugs;

[0030] Calculating the predicted drug resistance result and the actual drug resistance result of the sample isolate for the sample drug based on a binary cross entropy loss function to determine a loss value of the deep learning model;

[0031] If the loss value is greater than or equal to a preset threshold, the network parameters of the deep learning model are changed, and the changed deep learning model is continued to be iteratively trained. If the loss value is less than the preset threshold, the deep learning model is determined as the drug resistance prediction model.

[0032] In one possible implementation, the basic chemical properties of the atom include at least one of the following:

[0033] The electronegativity of the atom, the radius of the atom, the aromaticity of the atom, the hydrogen bond donor of the atom, and the hydrogen bond acceptor of the atom.

[0034] The present application also provides a device for predicting drug resistance, the device comprising:

[0035] A chemical molecular formula encoding module is used to input the chemical molecular structure of the drug into the molecular formula encoder of the drug resistance prediction model, perform molecular structure modeling on the chemical molecular structure, and output structural characteristic information of the drug;

[0036] A hierarchical encoding module is used to input the whole genome mutation information of the isolate into the hierarchical encoder of the drug resistance prediction model, perform encoding processing and attention mechanism processing on the whole genome mutation information, and output the encoding features of the whole genome mutation information;

[0037] A drug resistance determination module is used to input the structural feature information and the coding features into the classification network layer of the drug resistance prediction model for feature fusion and drug resistance classification processing, and output the drug resistance result of the isolate to the drug; wherein, the drug resistance prediction model is obtained by iterative training of a deep learning model.

[0038] An embodiment of the present application also provides an electronic device, comprising: a processor, a memory, and a bus, wherein the memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor and the memory communicate through the bus, and when the machine-readable instructions are executed by the processor, the steps of the drug resistance prediction method as described above are performed.

[0039] An embodiment of the present application further provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the above-mentioned drug resistance prediction method are executed.

[0040] The embodiment of the present application provides a drug resistance prediction method, device, electronic device and storage medium, the drug resistance prediction method comprising: inputting the chemical molecular structure of the drug into the molecular formula encoder of the drug resistance prediction model, performing molecular structure modeling processing on the chemical molecular structure, and outputting the structural feature information of the drug; inputting the whole genome mutation information of the isolate into the hierarchical encoder of the drug resistance prediction model, performing encoding processing and attention mechanism processing on the whole genome mutation information, and outputting the encoding features of the whole genome mutation information; inputting the structural feature information and the encoding features into the classification network layer of the drug resistance prediction model for feature fusion and drug resistance classification processing, and outputting the drug resistance result of the isolate for the drug; wherein, the drug resistance prediction model is obtained by iterative training of a deep learning model. By introducing the chemical molecular structure of the drug and learning the structural feature information of the drug, the structural feature information and the whole genome mutation representation are jointly modeled to capture more complex drug resistance patterns and accurately determine the drug resistance.

[0041] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without creative work.

[0043] Figure 1 A flowchart of a method for predicting drug resistance provided in an embodiment of the present application;

[0044] Figure 2 A schematic diagram of a method for predicting drug resistance provided in an embodiment of the present application;

[0045] Figure 3 This is one of the structural schematic diagrams of a drug resistance prediction device provided in an embodiment of the present application;

[0046] Figure 4 This is a second structural diagram of a drug resistance prediction device provided in an embodiment of the present application;

[0047] Figure 5 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0048] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. The components of the embodiments of the present application generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application for which protection is claimed, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, each other embodiment obtained by those skilled in the art without making creative work falls within the scope of protection of the present application.

[0049] First, the application scenarios to which this application is applicable are introduced. This application can be applied in the field of drug resistance prediction technology.

[0050] Research has found that while current drug resistance prediction methods have made some progress in improving the classification performance of specific drugs, they still face significant challenges in generalizing to new drugs or low-resource drugs that lack labeling. For example, when predicting whether MTB is resistant to anti-tuberculosis drugs, existing methods have a problem: if a new or rare drug lacks sufficient experimental data (i.e., labeled samples), the performance of these methods will be greatly reduced. This is because they rely on large amounts of known data to train the model. However, in the early stages of new drug development, such data is often scarce or even non-existent, which limits the identification of potential drug resistance risks and leads to technical problems such as inaccurate determination of drug resistance.

[0051] Based on this, an embodiment of the present application provides a drug resistance prediction method by introducing the chemical molecular structure of the drug and learning the structural characteristic information of the drug. The structural characteristic information is jointly modeled with the whole genome mutation representation to capture more complex resistance patterns and accurately determine the drug resistance.

[0052] See also Figure 1 , Figure 1 This is a flow chart of a method for predicting drug resistance provided in an embodiment of the present application. Figure 1 As shown in , the drug resistance prediction method provided in the embodiments of the present application includes:

[0053] S101: Inputting the chemical molecular structure of the drug into the molecular formula encoder of the drug resistance prediction model, performing molecular structure modeling processing on the chemical molecular structure, and outputting structural feature information of the drug.

[0054] In this step, the chemical molecular structure of the drug is input into the molecular formula encoder of the drug resistance prediction model, the chemical molecular structure is subjected to molecular structure modeling, and the structural characteristic information of the drug is output.

[0055] Here, the structural characteristic information of the drug integrates the chemical molecular structure and atomic properties of the drug.

[0056] In a possible implementation manner, performing molecular structure modeling on the chemical molecular structure and outputting structural feature information of the drug includes:

[0057] A: Encode the chemical molecular structure to generate the initial atomic features of the chemical molecular structure.

[0058] Here, the input chemical molecular structure is converted into a SMILES sequence and then encoded to obtain the initial atomic feature representation of the chemical molecular structure. i drugs , its atomic initial characteristics for:

[0059]

[0060] in, For feature extraction, For drugs molecular structure Sequence. The SMILES expression of the drug is split into a series of substructure tokens. Each token is further mapped into a high-dimensional vector to capture the contextual features of the local environment and functional groups in the molecular structure.

[0061] B: Parsing the chemical molecular structure into a molecular graph structure; wherein the nodes in the molecular graph structure are the atoms of the drug, and the edges are the chemical bonds of the drug.

[0062] In a possible implementation, the molecular graph structure is:

[0063]

[0064] Among them, V is the set of nodes, E is the set of edges consisting of chemical bonds between two atoms, is the molecular graph structure, For drugs The chemical molecular structure of sequence, For analysis tools.

[0065] Here, each edge Represents a (node) atom and The chemical bonds between.

[0066] B: Performing feature modeling based on the nodes, edges, and initial atomic features in the molecular graph structure to determine the structural feature information; wherein the structural feature information is a graph representation that integrates chemical molecular structure and atomic properties.

[0067] Here, feature modeling is performed based on the initial features of nodes, edges and atoms in the molecular graph structure to determine the structural feature information.

[0068] For example, for isoniazid, its SMILES representation is C1=CC=NC=C1C(=O)NN, corresponding to a chemical structure containing a pyridine aromatic ring (C1=CC=NC=C1) and a hydrazide group (C(=O)NN). To enhance the model's understanding of molecular topology, different types of chemical bonds are mapped to discrete edge weights: aromatic bonds (e.g., C–C, C–N) are assigned a value of 3, double bonds (C=O) are assigned a value of 2, and single bonds (e.g., C–N, N–N) are assigned a value of 1. These structural features are not only important in graph modeling but also closely related to the mechanism of action of anti-tuberculosis drugs. For example, mutations in KatG in MTB impair the in vivo activation of isoniazid, while mutations in the InhA binding site may interfere with the interaction of the carbonyl group (C=O) with the target protein, leading to drug ineffectiveness. By introducing different bond values ​​in the adjacency matrix, the model can characterize the functional characteristics of key functional groups at the structural level.

[0069] In a possible implementation manner, performing feature modeling based on the nodes, edges, and initial features of atoms in the molecular graph structure to determine the structural feature information includes:

[0070] a: The basic chemical properties of atoms are used as the initial feature vectors of the corresponding nodes, and the chemical bond type corresponding to each edge is used as the feature of the edge.

[0071] Here, in terms of node features, the present invention provides a Extract five basic chemical properties to form the initial feature vector ,like ,in, represents the electronegativity of the atom, represents the atomic radius, Indicates aromaticity, represents a hydrogen bond donor, Represents hydrogen bond receptors. The atomic-level features of all nodes are combined to form the atomic-level representation matrix of the drug molecular formula. .

[0072] For example, the basic chemical properties of the atom include at least one of the following: electronegativity of the atom, radius of the atom, aromaticity of the atom, hydrogen bond donor of the atom, and hydrogen bond acceptor of the atom.

[0073] Among them, each edge Represents atoms and The chemical bonds between them are extracted according to their type and used as edge features for modeling the molecular graph, as shown in the formula:

[0074]

[0075] in, Represents the type of chemical bond.

[0076] b: The initial feature vector of each node Feature fusion is performed with the initial atomic features to determine the target feature vector of each node.

[0077] Here, the initial feature vector of each node is With atomic initial characteristics Perform feature fusion to determine the target feature vector of each node .

[0078] c: For each node, aggregate the features of the edges between the node and its neighboring nodes and the target feature vectors of the neighboring nodes. Based on the aggregated features, update the target feature vector of the node to determine the updated feature vector of the node.

[0079] Here, for each node, the features of the edges between the node and the neighboring nodes and the target feature vectors of the neighboring nodes are aggregated, and the target feature vector of the node is updated according to the aggregated features to determine the updated feature vector of the node.

[0080] Among them, the neighbor node matrix of each node is determined according to the molecular graph structure, and the update of each node depends on the message aggregation of its neighbor nodes, as shown in the following formula:

[0081]

[0082] in, is a node The aggregated messages received, is the neighbor node, Used to integrate neighbor node features and edge type information Then, the node representation is updated by combining its own characteristics with the aggregated message, as shown in the following formula:

[0083]

[0084] in, is the update function, For nodes The updated feature vector of .

[0085] d: Processing the updated feature vector of each node based on a global average pooling function to determine the structural feature information.

[0086] Here, after propagation through the multi-layer graph neural network, the global average pooling function is used to integrate the features of all nodes to obtain the structural feature information of the overall representation of the molecular graph:

[0087]

[0088] in, For drugs i The structural feature information, is a node set, For nodes The updated feature vector of is the global average pooling function.

[0089] S102: Inputting the whole genome mutation information of the isolate into the hierarchical encoder of the drug resistance prediction model, performing encoding processing and attention mechanism processing on the whole genome mutation information, and outputting the encoding features of the whole genome mutation information.

[0090] In this step, the hierarchical encoder converts the isolate's whole genome sequence (e.g., in FASTA format) or mutation information (e.g., in VCF files) into a numerical matrix suitable for neural network processing. The bases (A, T, C, G) or mutation types (e.g., SNPs, Indels) in the genome are mapped into vector representations (e.g., one-hot encoding or embedding vectors). The first encoding layer captures local features. For example, convolutional layers detect mutation patterns within short fragments (e.g., the frequency distribution of single nucleotide polymorphisms (SNPs)). Intermediate encoding layers integrate contextual information. These layers may focus on longer-range dependencies, such as synergistic effects between mutation sites or mutation enrichment in functional regions of genes. High-level encoding generates a global representation. The final encoding layer may summarize the mutation characteristics of the entire genome, forming a low-dimensional but information-rich feature vector. The feature vector output by the high-level encoding is processed using an attention mechanism to obtain the encoded features of the whole-genome mutation information.

[0091] S103: Inputting the structural feature information and the coding features into the classification network layer of the drug resistance prediction model for feature fusion and drug resistance classification processing, and outputting the drug resistance result of the isolate to the drug; wherein, the drug resistance prediction model is obtained by iterative training of a deep learning model.

[0092] In this step, structural feature information and encoding features are modeled using a cross-attention mechanism to capture the relationship between the drug and the isolate in the resistance prediction summary. This fused joint representation is input into the resistance prediction classifier, which outputs the resistance prediction result for a given isolate under a specified drug.

[0093] In this application, to address the reliance on labeled drug-resistant samples, we considered constructing alternative supervisory signals using information about the drug itself, specifically its chemical molecular structure. A drug's molecular structure, as a direct reflection of its mechanism of action, carries atomic-level topological relationships and functional group characteristics. For example, among the various drugs used to treat tuberculosis, fluoroquinolones (such as Moxifloxacin and Levofloxacin) are not only highly similar in their pharmacological mechanisms but also in their molecular structures, containing a typical quinolone core structure and similar substituent groups. In contrast, aminoglycosides (such as Amikacin and Streptomycin) have a polysaccharide chain backbone and exhibit distinct topological configurations and physicochemical properties. This structural difference not only reflects the mechanistic basis for drug classification but also provides the model with molecular structural information that can be used to distinguish drug resistance patterns.

[0094] In a possible implementation, the inputting of the structural feature information and the coding feature into the classification network layer of the drug resistance prediction model for feature fusion and drug resistance classification processing, and outputting the drug resistance result of the isolate to the drug, includes:

[0095] (1): The structural feature information and the coding feature are spliced ​​together to determine the spliced ​​features.

[0096] Here, the features after splicing are determined by the following formula:

[0097]

[0098] in, is the feature after splicing, is the splicing function, It is the encoding feature of the whole genome mutation information.

[0099] (2): Based on the concatenated features and attention weights, the fusion features are determined.

[0100] (3): The fusion features are processed based on the feedforward neural sub-network layer and the activation function in the classification network layer, and the drug resistance result of the isolate to the drug is output.

[0101] Here, the fusion features The data is sent to the feedforward neural network layer (FFNN) and mapped to drug resistance results through the Sigmoid activation function. , which represents the drug resistance of the isolate under given drug conditions. The calculation method of this process is as follows:

[0102]

[0103] Among them, if P is 1, it means resistance, and if P is 0, it means sensitivity.

[0104] In one possible implementation, the drug resistance prediction model is determined by the following steps:

[0105] i: Input the sample chemical molecular structures of multiple sample drugs and the sample whole genome mutation information of the sample isolates into the deep learning model, and output the predicted drug resistance results of the sample isolates for the sample drugs.

[0106] Here, the deep learning model's processing of the sample's chemical molecular structure and the sample's whole genome mutation information is consistent with the processing of the above-mentioned drug resistance prediction model, and will not be repeated in this section.

[0107] For example, the sample dataset includes a collection of MTB isolates and anti-tuberculosis drug collection , where each isolate Expressed as , each Contains n mutation sites , the drug resistance prediction of MTB is modeled as a binary classification problem. For any isolate and drug pair , its drug resistance Indicates the resistance of isolate Z to drug D, where 0 indicates sensitive and 1 indicates resistant.

[0108] ii: Calculating the predicted drug resistance results and actual drug resistance results of the sample isolates to the sample drug based on the binary cross entropy loss function to determine the loss value of the deep learning model.

[0109] Here, the binary cross entropy loss function is used as the optimization target, and the loss function L Measuring the model's predictive value for predicting drug resistance outcomes and actual drug resistance results The deviation between them is shown in the following formula:

[0110]

[0111] iii: If the loss value is greater than or equal to a preset threshold, the network parameters of the deep learning model are changed, and the changed deep learning model is continued to be iteratively trained. If the loss value is less than the preset threshold, the deep learning model is determined as the drug resistance prediction model.

[0112] For further information, see Figure 2 , Figure 2 Schematic diagram of the drug resistance prediction method provided in the embodiment of this application. Figure 2 As shown, feature extraction is performed on the SMILES expression of the chemical molecular structure to determine the initial atomic features. The SMILES expression is parsed using the RDKit parsing tool to obtain a molecular graph structure. The molecular graph structure and initial atomic features are then integrated into the graph construction process. The contribution of different atoms to drug resistance is modeled through a graph attention mechanism to extract the structural characteristics of the drug. A hierarchical encoder is applied to the whole-genome mutation information of the isolate to obtain the encoding features of the whole-genome mutation information. The structural and encoding features are then input into the classification network layer for feature fusion and drug resistance classification, outputting the drug resistance results.

[0113] For example, after extensive experimental data analysis, the drug resistance prediction model proposed in this application demonstrated balanced performance across various metrics, achieving an AUC of 89.77%, the highest among all models and demonstrating superior classification capabilities. This result demonstrates that combining atomic initial features with molecular graph structure representation can effectively enhance the expressive power of drug molecular structural features, further improving the model's performance in drug resistance prediction tasks.

[0114] In this application, a more discriminative drug resistance prediction model is constructed by integrating drug molecular structure with genome-wide mutation information from isolates. This model converts the SMILES representation of a given drug into a molecular graph structure. The structural features are deeply modeled using the Generative Adversarial Analytical Test (GAT), and then integrated with the genome-wide mutation information through a cross-attention mechanism to ultimately achieve drug resistance classification. The specific implementation process involves splitting the drug's SMILES representation into a series of substructure tokens. Each token is further mapped into a high-dimensional vector to capture the contextual features of the local environment and functional groups within the molecular structure. Simultaneously, the SMILES is parsed into a molecular structure graph using the RDKit tool, constructing an adjacency matrix in which nodes represent atoms and edges represent covalent bonds between atoms. Edge attributes are further labeled as single bonds, double bonds, aromatic bonds, and other types, resulting in a graph with heterogeneous edge weights. Subsequently, the initial atomic features are integrated with the graph structure representation constructed by RDKit and fed into the Generative Adversarial Analytical Test (GAT). The graph attention mechanism is used to model the contribution of different atoms to drug resistance, thereby extracting the overall structural representation of the drug. Finally, the obtained drug expression and the whole genome expression of the isolate are combined through the information fusion layer to generate the combined expression of the isolate under the current drug condition. Finally, the fusion expression is sent to the drug resistance prediction classifier to obtain the prediction result.

[0115] The present application provides a method for predicting drug resistance, which includes: inputting the chemical molecular structure of the drug into the molecular formula encoder of the drug resistance prediction model, performing molecular structure modeling on the chemical molecular structure, and outputting the structural feature information of the drug; inputting the whole genome mutation information of the isolate into the hierarchical encoder of the drug resistance prediction model, performing encoding processing and attention mechanism processing on the whole genome mutation information, and outputting the encoding features of the whole genome mutation information; inputting the structural feature information and the encoding features into the classification network layer of the drug resistance prediction model for feature fusion and drug resistance classification processing, and outputting the drug resistance result of the isolate for the drug; wherein, the drug resistance prediction model is obtained by iterative training of a deep learning model. By introducing the chemical molecular structure of the drug and learning the structural feature information of the drug, the structural feature information and the whole genome mutation representation are jointly modeled to capture more complex drug resistance patterns and accurately determine the drug resistance.

[0116] See also Figure 3 、 Figure 4 , Figure 3 This is one of the structural schematic diagrams of a drug resistance prediction device provided in an embodiment of the present application; Figure 4 This is a second structural diagram of a drug resistance prediction device provided in an embodiment of the present application. Figure 3As shown in , the drug resistance prediction device 300 includes:

[0117] The chemical molecular formula encoding module 310 is used to input the chemical molecular structure of the drug into the molecular formula encoder of the drug resistance prediction model, perform molecular structure modeling on the chemical molecular structure, and output structural characteristic information of the drug;

[0118] A hierarchical encoding module 320 is configured to input the whole genome mutation information of the isolate into the hierarchical encoder of the drug resistance prediction model, perform encoding and attention mechanism processing on the whole genome mutation information, and output encoding features of the whole genome mutation information;

[0119] The drug resistance determination module 330 is used to input the structural feature information and the coding features into the classification network layer of the drug resistance prediction model for feature fusion and drug resistance classification processing, and output the drug resistance result of the isolate to the drug; wherein, the drug resistance prediction model is obtained by iterative training of a deep learning model.

[0120] Furthermore, when the chemical formula encoding module 310 is used to perform molecular structure modeling on the chemical molecular structure and output structural characteristic information of the drug, the chemical formula encoding module 310 is specifically used to:

[0121] Encoding the chemical molecular structure to generate an atomic initial feature of the drug;

[0122] Parsing the chemical molecular structure into a molecular graph structure; wherein the nodes in the molecular graph structure are atoms of the drug and the edges are chemical bonds of the drug;

[0123] Feature modeling is performed based on the nodes, edges and initial atomic features in the molecular graph structure to determine the structural feature information; wherein the structural feature information is a graph representation that integrates the chemical molecular structure, atomic properties and initial atomic features.

[0124] Furthermore, when the chemical formula encoding module 310 is used to perform feature modeling based on the nodes, edges, and initial atomic features in the molecular graph structure to determine the structural feature information, the chemical formula encoding module 310 is specifically used to:

[0125] The basic chemical properties of atoms are used as the initial feature vectors of the corresponding nodes, and the chemical bond type corresponding to each edge is used as the feature of the edge;

[0126] Performing feature fusion on the initial feature vector of each node and the initial feature of the atom to determine the target feature vector of each node;

[0127] For each node, the features of the edges between the node and its neighboring nodes and the target feature vectors of the neighboring nodes are aggregated, and the target feature vector of the node is updated based on the aggregated features to determine the updated feature vector of the node;

[0128] The updated feature vector of each node is processed based on a global average pooling function to determine the structural feature information.

[0129] Furthermore, the drug resistance determination module 330 is used to input the structural feature information and the coding feature into the classification network layer of the drug resistance prediction model to perform feature fusion and drug resistance classification processing, and output the drug resistance result of the isolate for the drug. The drug resistance determination module 330 is specifically used to:

[0130] Performing feature splicing on the structural feature information and the coding feature to determine the spliced ​​feature;

[0131] Determine the fusion features based on the spliced ​​features and attention weights;

[0132] The fusion features are processed based on the feedforward neural sub-network layer and the activation function in the classification network layer, and the drug resistance result of the isolated strain to the drug is output.

[0133] Further, such as Figure 4 As shown, the drug resistance prediction device 300 further includes a model training module 340. The model training module 340 determines the drug resistance prediction model through the following steps:

[0134] Inputting sample chemical molecular structures of multiple sample drugs and sample whole genome mutation information of sample isolates into the deep learning model, and outputting predicted drug resistance results of the sample isolates for the sample drugs;

[0135] Calculating the predicted drug resistance result and the actual drug resistance result of the sample isolate for the sample drug based on a binary cross entropy loss function to determine a loss value of the deep learning model;

[0136] If the loss value is greater than or equal to a preset threshold, the network parameters of the deep learning model are changed, and the changed deep learning model is continued to be iteratively trained. If the loss value is less than the preset threshold, the deep learning model is determined as the drug resistance prediction model.

[0137] The embodiment of the present application provides a drug resistance prediction device, which includes: a chemical formula encoding module for inputting the chemical molecular structure of the drug into the molecular formula encoder of the drug resistance prediction model, performing molecular structure modeling on the chemical molecular structure, and outputting the structural feature information of the drug; a hierarchical encoding module for inputting the whole genome mutation information of the isolate into the hierarchical encoder of the drug resistance prediction model, performing encoding processing and attention mechanism processing on the whole genome mutation information, and outputting the encoding features of the whole genome mutation information; a drug resistance determination module for inputting the structural feature information and the encoding features into the classification network layer of the drug resistance prediction model for feature fusion and drug resistance classification processing, and outputting the drug resistance result of the isolate for the drug; wherein the drug resistance prediction model is obtained by iterative training of a deep learning model. By introducing the chemical molecular structure of the drug and learning the structural feature information of the drug, the structural feature information is jointly modeled with the whole genome mutation representation to capture more complex drug resistance patterns and accurately determine the drug resistance. By introducing the chemical molecular structure of the drug and learning the structural feature information of the drug, the structural feature information is jointly modeled with the whole genome mutation representation to capture more complex resistance patterns and accurately determine the drug resistance.

[0138] See also Figure 5 , Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. Figure 5 As shown in FIG, the electronic device 500 includes a processor 510, a memory 520 and a bus 530.

[0139] The memory 520 stores machine-readable instructions executable by the processor 510. When the electronic device 500 is running, the processor 510 communicates with the memory 520 via the bus 530. When the machine-readable instructions are executed by the processor 510, the above-mentioned Figure 1 as well as Figure 2 The specific implementation of the steps of the drug resistance prediction method in the method embodiment shown can be found in the method embodiment and will not be repeated here.

[0140] The embodiment of the present application also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the computer program can execute the above-mentioned Figure 1 as well as Figure 2 The specific implementation of the steps of the drug resistance prediction method in the method embodiment shown can be found in the method embodiment and will not be repeated here.

[0141] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0142] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. There may be other division methods in actual implementation. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed may be through some communication interface, indirect coupling or communication connection of devices or units, which may be electrical, mechanical or other forms.

[0143] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0144] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0145] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium that is executable by a processor. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes various media that can store program code, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0146] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present application, which are used to illustrate the technical solutions of the present application, rather than to limit them. The scope of protection of the present application is not limited thereto. Although the present application has been described in detail with reference to the above-mentioned embodiments, those skilled in the art should understand that any person skilled in the art can modify or easily conceive of changes to the technical solutions described in the above-mentioned embodiments within the technical scope disclosed in the present application, or perform equivalent replacements for some of the technical features thereof. These modifications, changes, or replacements do not deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A method for predicting drug resistance, characterized in that: The drug resistance prediction method comprises: Inputting the chemical molecular structure of the drug into the molecular formula encoder of the drug resistance prediction model, performing molecular structure modeling processing on the chemical molecular structure, and outputting structural characteristic information of the drug; The performing molecular structure modeling on the chemical molecular structure and outputting structural characteristic information of the drug includes: Encoding the chemical molecular structure to generate initial atomic features of the chemical molecular structure; Parsing the chemical molecular structure into a molecular graph structure; wherein the nodes in the molecular graph structure are atoms of the drug and the edges are chemical bonds of the drug; Performing feature modeling based on the nodes, edges, and initial atomic features in the molecular graph structure to determine the structural feature information; wherein the structural feature information is a graph representation that integrates the chemical molecular structure, atomic properties, and initial atomic features; The performing feature modeling based on the nodes, the edges, and the initial features of the atoms in the molecular graph structure to determine the structural feature information includes: The basic chemical properties of atoms are used as the initial feature vectors of the corresponding nodes, and the chemical bond type corresponding to each edge is used as the feature of the edge; Performing feature fusion on the initial feature vector of each node and the initial feature of the atom to determine the target feature vector of each node; For each node, the features of the edges between the node and its neighboring nodes and the target feature vectors of the neighboring nodes are aggregated, and the target feature vector of the node is updated based on the aggregated features to determine the updated feature vector of the node; Processing the updated feature vector of each node based on a global average pooling function to determine the structural feature information; Inputting the whole genome mutation information of the isolate into the hierarchical encoder of the drug resistance prediction model, performing encoding processing and attention mechanism processing on the whole genome mutation information, and outputting the encoding features of the whole genome mutation information; The structural feature information and the coding features are input into the classification network layer of the drug resistance prediction model for feature fusion and drug resistance classification processing, and the drug resistance result of the isolate to the drug is output; wherein, the drug resistance prediction model is obtained by iterative training of a deep learning model.

2. The drug resistance prediction method according to claim 1, wherein The molecular graph structure is: Among them, V is the set of nodes, E is the set of edges consisting of chemical bonds between two atoms, is the molecular graph structure, For drugs The chemical molecular structure of sequence, For analysis tools.

3. The drug resistance prediction method according to claim 1, wherein The step of inputting the structural feature information and the coding feature into the classification network layer of the drug resistance prediction model for feature fusion and drug resistance classification processing, and outputting the drug resistance result of the isolate to the drug, includes: Performing feature splicing on the structural feature information and the coding feature to determine the spliced ​​feature; Based on the concatenated features and attention weights, the fusion features are determined; The fusion features are processed based on the feedforward neural sub-network layer and the activation function in the classification network layer, and the drug resistance result of the isolated strain to the drug is output.

4. The drug resistance prediction method according to claim 1, wherein The drug resistance prediction model is determined by the following steps: Inputting sample chemical molecular structures of multiple sample drugs and sample whole genome mutation information of sample isolates into the deep learning model, and outputting predicted drug resistance results of the sample isolates for the sample drugs; Calculating the predicted drug resistance result and the actual drug resistance result of the sample isolate for the sample drug based on a binary cross entropy loss function to determine a loss value of the deep learning model; If the loss value is greater than or equal to a preset threshold, the network parameters of the deep learning model are changed, and the changed deep learning model is continued to be iteratively trained. If the loss value is less than the preset threshold, the deep learning model is determined as the drug resistance prediction model.

5. The drug resistance prediction method according to claim 1, wherein The basic chemical properties of the atoms include at least one of the following: The electronegativity of the atom, the radius of the atom, the aromaticity of the atom, the hydrogen bond donor of the atom, and the hydrogen bond acceptor of the atom.

6. A drug resistance prediction device, characterized in that: The drug resistance prediction device comprises: A chemical molecular formula encoding module is used to input the chemical molecular structure of the drug into the molecular formula encoder of the drug resistance prediction model, perform molecular structure modeling on the chemical molecular structure, and output structural characteristic information of the drug; The chemical molecular formula encoding module is used to perform molecular structure modeling on the chemical molecular structure and output the structural characteristic information of the drug: Encoding the chemical molecular structure to generate initial atomic features of the chemical molecular structure; Parsing the chemical molecular structure into a molecular graph structure; wherein the nodes in the molecular graph structure are atoms of the drug and the edges are chemical bonds of the drug; Performing feature modeling based on the nodes, edges, and initial atomic features in the molecular graph structure to determine the structural feature information; wherein the structural feature information is a graph representation that integrates the chemical molecular structure, atomic properties, and initial atomic features; The chemical molecular formula encoding module is used to perform feature modeling based on the nodes, edges, and initial atomic features in the molecular graph structure to determine the structural feature information: The basic chemical properties of atoms are used as the initial feature vectors of the corresponding nodes, and the chemical bond type corresponding to each edge is used as the feature of the edge; Performing feature fusion on the initial feature vector of each node and the initial feature of the atom to determine the target feature vector of each node; For each node, the features of the edges between the node and its neighboring nodes and the target feature vectors of the neighboring nodes are aggregated, and the target feature vector of the node is updated based on the aggregated features to determine the updated feature vector of the node; Processing the updated feature vector of each node based on a global average pooling function to determine the structural feature information; A hierarchical encoding module is used to input the whole genome mutation information of the isolate into the hierarchical encoder of the drug resistance prediction model, perform encoding processing and attention mechanism processing on the whole genome mutation information, and output the encoding features of the whole genome mutation information; A drug resistance determination module is used to input the structural feature information and the coding features into the classification network layer of the drug resistance prediction model for feature fusion and drug resistance classification processing, and output the drug resistance result of the isolate to the drug; wherein, the drug resistance prediction model is obtained by iterative training of a deep learning model.

7. An electronic device, characterized in that: include: A processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor and the memory communicate through the bus, and the machine-readable instructions are executed by the processor to execute the steps of the drug resistance prediction method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the drug resistance prediction method according to any one of claims 1 to 5 are executed.

Images