Drug-target interaction prediction method
By constructing a multi-view similarity network and the Interformer-CMPNN architecture, the problems of single feature extraction and limited model architecture in drug-target interaction prediction in existing technologies are solved, achieving efficient and accurate prediction of drug-target interactions and improving drug development efficiency.
Patent Information
- Application Number
- CN202511654460.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-12
- Publication Date
- 2026-02-10
AI Technical Summary
Existing machine learning-based drug-target interaction prediction methods suffer from limitations such as single feature extraction, neglect of multi-perspective biological data correlation, and limited model architecture, failing to fully capture the bidirectional dynamic influence between drugs and targets.
We construct a five-dimensional drug similarity network and a four-dimensional target similarity network, integrate heterogeneous information through a weighted fusion strategy, design an Interformer-CMPNN hybrid architecture, and use the Transformer self-attention mechanism to capture the bidirectional influence between drugs and targets and extract interactive features.
It significantly improves the accuracy and interpretability of drug-target interaction prediction, providing an efficient computational tool for drug development.
Smart Images
Figure CN121506233A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the field of artificial intelligence and drug technology, and in particular to a drug-target interaction prediction method. BACKGROUND
[0002] Drug-target interaction analysis plays an irreplaceable role as an important part of drug discovery. In order to find safe and effective drugs, traditional drug-target interaction (DTI) analysis often needs to test thousands of compounds, which requires a lot of manpower, material resources and time cost, and has a high risk of test failure. In recent years, computer-aided drug design has attracted more and more attention from drug researchers. Using artificial intelligence and other technologies, drug-target pairs (DTPs) that may interact are screened from massive drug and protein related data for further testing and verification by drug scientists. This method not only can reduce the waste of manpower and material resources in the drug discovery process, but also can greatly shorten the time of drug discovery.
[0003] The existing DTI prediction method based on machine learning has the problem of single feature extraction, only using drug chemical structure or target sequence information, ignoring the correlation of multi-perspective biological data, and the model architecture is limited, and the bidirectional dynamic influence between drugs and targets is not fully captured. SUMMARY
[0004] The present application provides a drug-target interaction prediction method to solve the problems raised in the background.
[0005] In order to achieve the above purpose, the present application adopts the following technical scheme:
[0006] A drug-target interaction prediction method, comprising the following steps:
[0007] S1, constructing a five-dimensional drug similarity network and a four-dimensional target similarity network;
[0008] S2, generating a drug similarity matrix and a target similarity matrix by a weighted fusion strategy ;
[0009] S3, extracting drug-target bidirectional interaction features using Interformer-CMPNN architecture;
[0010] S4, inputting the fusion features into a multilayer perceptron to predict potential drug-target interactions, and verifying the performance through ten-fold cross-validation and external data set.
[0011] As a further improved scheme of the technical solution: the constructed drug five-dimensional similar network includes drug-drug interaction, drug-disease correlation, drug-side effect correlation, drug-protein interaction and chemical structure similarity.
[0012] As a further improved scheme of the technical solution: the constructed target four-dimensional similar network includes protein-protein interaction, protein-disease correlation, drug-protein interaction and gene sequence similarity.
[0013] As a further improved scheme of the technical solution: in the weighted fusion strategy, the drug similarity matrix calculation formula is: ;
[0014] Among them, is the chemical structure similarity matrix, is the Gaussian kernel similarity matrix, is the drug-drug interaction binary matrix, is the drug-side effect correlation similarity matrix, is the drug-target interaction similarity matrix, , , , , is the weight coefficient.
[0015] As a further improved scheme of the technical solution: in the weighted fusion strategy, the target similarity matrix calculation formula is: , wherein is the protein-protein interaction similarity matrix, is the Gaussian kernel similarity matrix, is the drug-target interaction similarity matrix, is the gene sequence similarity matrix, , , , is the weight coefficient.
[0016] As a further improved scheme of the technical solution: the Interformer-CMPNN architecture includes:
[0017] Two parallel Transformer decoders, respectively processing drug and target features;
[0018] An interactive attention layer that receives external information from the other decoder;
[0019] An Inter-CMPNN module that integrates drug-target interaction information.
[0020] As a further improvement to this technical solution: the formula for calculating the Gaussian kernel similarity matrix is: ;in This is the correlation vector between the drug and the target. This is the kernel width parameter.
[0021] As a further improvement to this technical solution: the calculation method of the interactive attention layer is as follows:
[0022] ;in For drug lookup matrix, For the target bond matrix, For value matrices, For the attention dimension.
[0023] The present invention also includes a computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements any of the above-described drug-target interaction prediction methods.
[0024] Compared with the prior art, the beneficial effects of the present invention are:
[0025] This invention constructs a five-dimensional drug similarity network and a four-dimensional target similarity network, integrates heterogeneous information through a weighted fusion strategy, and designs an Interformer-CMPNN hybrid architecture. It utilizes the Transformer self-attention mechanism to capture the bidirectional influence between drugs and targets, and extracts interaction features through an improved message passing neural network. This significantly improves the accuracy and interpretability of DTI prediction, providing an efficient computational tool for drug development.
[0026] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it according to the contents of the specification, the preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings. Specific embodiments of the present invention are given in detail below with reference to the accompanying drawings. Attached Figure Description
[0027] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:
[0028] Figure 1 This is a schematic diagram illustrating the steps of a drug-target interaction prediction method proposed in this invention. Detailed Implementation
[0029] The principles and features of the present invention are described below with reference to the accompanying drawings. The examples given are for illustrative purposes only and are not intended to limit the scope of the invention. The invention is described more specifically in the following paragraphs by way of example with reference to the accompanying drawings. It should be noted that the drawings are in a very simplified form and use non-precise proportions, and are only used to facilitate and clarify the illustration of the embodiments of the present invention.
[0030] In this embodiment of the invention, a method for predicting drug-target interactions includes the following steps:
[0031] Step 1, Multi-view Similarity Network Construction: Integrating a five-dimensional drug similarity network and a four-dimensional target similarity network to capture the complex relationships between drugs and targets. Specifically:
[0032] The five-dimensional similarity network of drugs is as follows:
[0033] Drug-Drug Interactions (DDI): A binary matrix is constructed based on known DDI data, with parameters representing drug pairs ( , The interaction flag (1 for interaction, 0 for no interaction).
[0034] Drug-Disease Association (DGA): Gaussian kernel similarity is calculated based on the DisGeNET database, with parameters being drugs. With disease correlation score ,formula: , where σ is the kernel width parameter, determined using the median heuristic.
[0035] Drug-Side Effect Association (DSE): A similarity matrix is constructed based on the SIDER database, with parameters being drugs. and side effects correlation vector .
[0036] Drug-protein interaction (DPI): A binary matrix was constructed based on the DrugBank database, with parameters representing drugs. With the target Interaction markers.
[0037] Chemical Structure Similarity (CHEM): Tanimoto similarity was calculated based on ECFP fingerprinting, with parameters being the drug... and Chemical structure fingerprint vector and ,formula: .
[0038] The target four-dimensional similarity network is as follows:
[0039] Protein-protein interactions (PPI): A similarity matrix was constructed based on the STRING database, with parameters representing the target. and The confidence level of the interaction.
[0040] Protein-Disease Association (PTA): Gaussian kernel similarity was calculated based on the DisGeNET database, with the parameter being the target. With disease correlation score .
[0041] Drug-protein interaction (DPI): Shares parameters with DPIs in the drug network.
[0042] Gene sequence similarity (SEQ): Sequence alignment score is calculated based on the Smith-Waterman algorithm, with the parameter being the target. and Sequence alignment score .
[0043] Step 2, Weighted Fusion Strategy: By dynamically allocating weights, multi-view similar networks are fused to generate enhanced feature representations;
[0044] Specifically, drug similarity matrix fusion: A five-dimensional drug similarity network is fused through dynamic weight allocation, using the formula:
[0045] ;
[0046] in, This is a chemical structure similarity matrix. The Gaussian kernel similarity matrix is... This is a binary matrix representing drug-drug interactions. This is a drug-side effect association similarity matrix. This is a drug-target interaction similarity matrix. , , , , The weighting coefficients are determined through grid search optimization.
[0047] Target similarity matrix fusion: Similarly, fuse the four-dimensional target similarity network, formula: ,in , , , The weighting coefficients are determined through grid search optimization.
[0048] Step 3, Deep Interactive Attention Mechanism: Using the Interformer-CMPNN architecture, model the bidirectional interaction between the drug and the target, and extract discriminative features;
[0049] The Interformer-CMPNN architecture includes: two parallel Transformer decoders, which process drug and target features respectively; an interactive attention layer, which receives external information from the other decoder; and an Inter-CMPNN module, which integrates drug-target interaction information.
[0050] The specific formula for the interactive attention layer is: ;in For drug lookup matrix, For the target bond matrix, For value matrices, For the attention dimension (set to 64).
[0051] Inter-CMPNN module: Parameters are drug atom feature matrices and target residue feature matrix Update node characteristics through message passing, formula: Msg is the message function, and Update is the update function.
[0052] Step 4, Prediction and Validation: Input the fused features into a multilayer perceptron (MLP) to predict the latent DTI, and validate the performance using 10-fold cross-validation against an external dataset. Specifically:
[0053] The prediction model is as follows: The fused features are input into the MLP, and the output is the DTI prediction probability. The formula is: ;in For the sigmoid function, , This is the weight matrix. This is a bias term.
[0054] The validation method is: 10-fold cross-validation: the dataset is divided into 10 parts, and 9 parts are used for training and 1 part for testing in turn. The average AUC and AUPR are calculated.
[0055] External dataset validation: Introduce external datasets such as BindingDB and Davis to validate the model's generalization ability.
[0056] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Those skilled in the art can readily implement the present invention based on the accompanying drawings and the above description. However, any modifications, alterations, or variations made by those skilled in the art without departing from the scope of the present invention, utilizing the disclosed technical content, are equivalent embodiments of the present invention. Furthermore, any modifications, alterations, or variations made to the above embodiments based on the essential technology of the present invention are still within the protection scope of the present invention.
Claims
1. A method for predicting drug-target interactions, characterized in that, Includes the following steps: S1, construct a five-dimensional drug similarity network and a four-dimensional target similarity network; S2 generates a drug similarity matrix through a weighted fusion strategy. Similarity matrix with target ; S3 utilizes the Interformer-CMPNN architecture to extract bidirectional drug-target interaction features; S4 inputs fused features into a multilayer perceptron to predict potential drug-target interactions, and validates its performance against an external dataset using 10-fold cross-validation.
2. The method for predicting drug-target interactions according to claim 1, characterized in that, The constructed drug five-dimensional similarity network includes drug-drug interactions, drug-disease associations, drug-side effect associations, drug-protein interactions, and chemical structure phases.
3. The method for predicting drug-target interactions according to claim 2, characterized in that, Construct a four-dimensional target similarity network, including protein-protein interactions, protein-disease associations, drug-protein interactions, and gene sequence similarity.
4. The method for predicting drug-target interactions according to claim 3, characterized in that, In the weighted fusion strategy, the formula for calculating the drug similarity matrix is: ; in, This is a chemical structure similarity matrix. The Gaussian kernel similarity matrix is... This is a binary matrix representing drug-drug interactions. This is a drug-side effect association similarity matrix. This is a drug-target interaction similarity matrix. , , , , These are the weighting coefficients.
5. The method for predicting drug-target interactions according to claim 4, characterized in that, In the weighted fusion strategy, the formula for calculating the target similarity matrix is: ,in, This is a protein-protein interaction similarity matrix. The Gaussian kernel similarity matrix is... This is a drug-target interaction similarity matrix. This is a gene sequence similarity matrix. , , , These are the weighting coefficients.
6. The method for predicting drug-target interactions according to claim 5, characterized in that, The Interformer-CMPNN architecture includes: Two parallel Transformer decoders process drug and target features respectively; An interactive attention layer receives external information from another decoder; The Inter-CMPNN module integrates drug-target interaction information.
7. The method for predicting drug-target interactions according to claim 6, characterized in that, The formula for calculating the Gaussian kernel similarity matrix is: ;in This is the correlation vector between the drug and the target. This is the kernel width parameter.
8. The method for predicting drug-target interactions according to claim 7, characterized in that, The calculation method for the interactive attention layer is as follows: ;in For drug lookup matrix, For the target bond matrix, For value matrices, For the attention dimension.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the drug-target interaction prediction method as described in any one of claims 1-8.