一种蛋白质相互作用位点预测方法及系统
By combining the local structure and global interaction modules of the isomorphic graph neural network and the spatially aware Transformer, and using the centroid coordinates of the residue side chains and the cosine of the angle to correct the attention matrix, the problem of ignoring local and global structural information in existing methods is solved, and the prediction accuracy of protein interaction sites is improved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- JIANGNAN UNIV
- Filing Date
- 2025-07-24
- Publication Date
- 2026-07-17
AI Technical Summary
Existing graph neural network-based protein-protein interaction site prediction methods neglect the intrinsic connection between the local structural features of residues and the global structural information of proteins, as well as the spatial relationship between residue nodes, resulting in poor prediction accuracy.
A local structure and global interaction module cascaded by residual-connected isomorphic graph neural network and spatially aware Transformer is used. The attention matrix is modified by the centroid coordinates of residues to side chains and the cosine of the included angle through a multi-head attention mechanism. Combined with a regularized Laplacian matrix and a multi-step random walk matrix, high-order features of proteins are extracted.
By effectively integrating local and global structural information, the prediction accuracy of protein-protein interaction sites is improved, the ability to characterize protein geometric features is enhanced, and the prediction precision is increased.
Smart Images
Figure CN120877854B_ABST
Abstract
Citation Information
Patent Citations
Protein binding site prediction method, system, medium, equipment and product
CN118522346A
Protein interaction site prediction method based on local-global feature fusion
CN119964639A