A Structure-Based Method for Prediction of Protein-Protein Interaction Sites Through Combining Local and Global Features

The identification of protein-protein interaction(PPI) sites is critical for drug discovery and analysis of protein function. So far, numerous computational methods based on artificial intelligence have been developed to accelerate the identification of PPI sites. However, those methods learned only sequential features or local structural features, leading to low predictive performance. Therefore, it is urgently necessary to develop a new method to improve the performance of predicting PPI sites. Herein, a new structure-based method for predicting the protein-protein interaction sites through combing the local and global encoder, SLGPPIS, was therefore proposed. Specifically, SLGPPIS utilizes the local and global encoder to extract the residue feature respectively and concatenating them to predict the PPI sites. Compared with other existing methods, SLGPPIS exhibited superior performance on prediction of PPI sites. Moreover, a visual interpretative experiment was conducted to systematically investigate the prediction process of SLGPPIS.