Compound-protein interaction prediction method fusing gating mechanism and contrast learning

Abstract

The invention provides a compound-protein interaction prediction method fusing a gating mechanism and contrast learning, which comprises the following steps: firstly, carrying out feature extraction on an amino acid sequence and a compound molecule through a coding layer, segmenting the amino acid sequence by adopting an n-gram method so as to retain local sequence features, and carrying out feature extraction on the compound molecule; the method comprises the following steps of: extracting a sequence of a compound molecule, capturing orderliness and long-term dependency in the sequence through a long-short-term memory network (LSTM) so as to obtain richer sequence representation, extracting atomic characteristics and molecular structure information of the compound molecule by utilizing a graph neural network (GNN), and fully capturing a topological relation and a local structure in a molecular graph; next, the model fuses the extracted features through a gating mechanism, redundant information is effectively filtered out, and key features are reserved, so that the precision of feature expression is enhanced; on the basis, the Contrast Learning is introduced, and the distinguishing ability of the model is further improved through the Contrast Learning of similar and non-similar samples, so that the interaction between the compound and the protein can be identified more accurately; finally, the model splices the features of the compound and the protein through a decoding layer, and accurate prediction of the interaction is achieved.

Description

Technical Field

[0001] The present invention relates to the field of bioinformatics and deep learning technology, and specifically to a compound-protein interaction (CPI) prediction model that integrates gating mechanism and contrastive learning. The model aims to solve the limitations of existing CPI prediction methods in feature extraction and information fusion, and improves the accuracy and reliability of CPI prediction by introducing gating mechanism and contrastive learning technology, thereby accelerating the process of new drug research and development and reducing drug research and development costs. The technical field involved in the present invention covers the computational prediction of compound-protein interactions, including but not limited to biochemical experimental methods, structure-based computational methods, and machine learning-based prediction methods. On this basis, the present invention proposes a new computational model that combines the gating mechanism and contrastive learning technology in deep learning, which can more effectively extract and fuse the characteristic information of compounds and proteins, and achieve accurate prediction of CPI. Background Art

[0002] The statements in this section only provide background information related to the present invention and do not necessarily constitute prior art. In the field of biomedicine and drug development, the prediction of compound-protein interaction (CPI) is a key link in the drug discovery process. Accurate prediction of CPI helps researchers to quickly screen potential drug candidate molecules, thereby accelerating the process of new drug development and improving the success rate of drug development. However, there are many challenges in traditional CPI prediction methods. On the one hand, although biochemical experimental methods can provide accurate CPI data, the experimental process is cumbersome, time-consuming and costly, and it is difficult to meet the needs of large-scale drug screening. On the other hand, although structure-based computational methods can predict the binding sites of compounds and proteins, they are limited by the difficulty of obtaining the three-dimensional structure of proteins and the complexity of calculations, and their scope of application is limited. In recent years, with the continuous development of machine learning technology, CPI prediction methods based on machine learning have gradually emerged. These methods extract characteristic information of compounds and proteins and use machine learning algorithms to establish prediction models to achieve the prediction of CPI. However, existing CPI prediction methods based on machine learning still face some technical difficulties. For example, how to effectively extract and fuse the characteristic information of compounds and proteins to improve the accuracy and reliability of predictions; how to overcome problems such as data sparsity and imbalance to improve the generalization ability of the model. Summary of the invention

[0003] In order to solve the above problems, the present invention proposes an innovative CPI prediction model, which integrates the gating mechanism and contrastive learning technology, aiming to improve the accuracy and reliability of CPI prediction and accelerate the process of new drug research and development. The core of the present invention is to construct a CPI prediction model based on deep learning, which consists of the following key parts: Feature extraction module: Using deep learning technology, automatically extract features from the structural information of compounds and proteins. These features include but are not limited to the chemical structure features of compounds, the three-dimensional structural features of proteins, and the interaction interface features between them. Gating mechanism module: On the basis of feature extraction, a gating mechanism is introduced to dynamically adjust the fusion ratio of different feature information. The gating mechanism can intelligently select important features for fusion according to contextual information, thereby avoiding the interference of redundant information and improving the prediction performance of the model. Contrastive learning module: In order to further improve the generalization ability of the model, the present invention introduces contrastive learning technology. By comparing the interaction patterns between different compounds and proteins, the model can learn more robust and generalized feature representations, thereby enhancing the prediction ability of unknown CPI. Prediction output module: Based on the results of the above feature extraction, gating mechanism and contrastive learning, the model finally outputs the prediction results of CPI. These results may include information such as whether the compound interacts with the protein, the strength of the interaction, and possible sites of action. Through the CPI prediction model of the present invention, researchers can screen potential drug candidate molecules more quickly and accurately, thereby significantly reducing the cost and time cost of new drug research and development. In addition, the model also has broad application prospects, including but not limited to the fields of drug repositioning, drug side effect prediction, and biomarker discovery. Compared with the prior art, the beneficial effects of the present invention are: by fusing the gating mechanism with the comparative learning technology, the present invention can more effectively extract and fuse the characteristic information of compounds and proteins, thereby significantly improving the accuracy of CPI prediction. This improvement is of great significance for compound screening and drug optimization in the drug development process, and helps scientific researchers find potential drug candidate molecules more quickly, thereby significantly reducing the cost and time cost of new drug research and development. The advantages of the present invention and the advantages of additional aspects will be described in detail in the following specific implementation methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0004] Figure 1 Flowchart of computational model for predicting interactions between compounds and proteins.

Claims

1. A compound-protein interaction prediction method integrating gating mechanism and contrastive learning, characterized in that: The process includes the following: Feature extraction of amino acid sequences and compound molecules is performed through the coding layer; The extracted features are fused using the gating mechanism to effectively filter out redundant information and retain key features; Introducing contrastive learning to improve the model's ability to distinguish between similar and dissimilar samples; And the prediction results of compound-protein interaction are output through the decoding layer.

2. A compound-protein interaction prediction method integrating gating mechanism and comparative learning as claimed in claim 1, characterized in that: When the encoding layer extracts features from the amino acid sequence, it segments it using the n-gram method and captures the orderliness and long-term dependency in the sequence through a long short-term memory network (LSTM).

3. The compound-protein interaction prediction method integrating gating mechanism and comparative learning as claimed in claim 1, characterized in that: When the encoding layer extracts features from compound molecules, it uses a graph neural network (GNN) to extract atomic features and molecular structure information, fully capturing the topological relationships and local structures in the molecular graph.

4. The compound-protein interaction prediction method integrating gating mechanism and comparative learning as claimed in claim 1, characterized in that: The gating mechanism compresses the output to the interval [0, 1] through the Sigmoid activation function to indicate the importance of each feature, and screens out important features through element-by-element product (Hadamard product) while suppressing unnecessary noise features.

5. The compound-protein interaction prediction method integrating gating mechanism and comparative learning as claimed in claim 1, characterized in that: The contrastive learning measures the similarity between features by calculating the cosine similarity between anchor samples and negative samples, and optimizes the contrastive loss function so that the model can better distinguish different features during the learning process.

6. The compound-protein interaction prediction method integrating gating mechanism and comparative learning as claimed in claim 1, characterized in that: The decoding layer performs the final prediction task on the extracted and processed protein and compound features. The fully connected layer is used to fuse the processed features, and finally the predicted probability distribution is output through the Softmax activation function, the features are mapped to the label space, and the prediction results of drug-protein interaction are output.

Images