Abstract
BACKGROUND: Protein-protein interactions (PPIs) play a crucial role in numerous biological processes. Accurate identification of protein-protein interaction sites is critical for a comprehensive understanding of protein functions and pathological mechanisms. However, conventional experimental approaches for detecting PPIs are often time-consuming and labor-intensive, thereby motivating the development of efficient computational methods to identify PPI sites. RESULTS: In this work, we propose a novel graph neural network-based method (called MGMA-PPIS) to predict PPI sites by adopting multiview graph embedding and multiscale attention fusion. MGMA-PPIS integrates global node features extracted by an equivariant graph neural network and multiscale local node features extracted by an edge graph attention network across different neighborhood scales, thereby constructing a multiview graph feature representation. Then, a multiscale attention network is employed to perform deep feature fusion across multiple scales for achieving high-precision prediction of PPI sites. CONCLUSIONS: Experimental results on benchmark datasets show that our MGMA-PPIS outperforms other state-of-the-art methods, and it can effectively predict PPI sites.