SSIF-Affinity: Multimodal Deep Learning of Sequence-Structure Features for Precise Protein–Protein Binding Affinity Prediction

作者：Xinyi Xu, Haotian Zhang, Qi Liu, Qian Cheng, Yanbei Guo, Yang Wei, Xueli Chen · 发表于：Journal of Chemical Information and Modeling · 年份：2025 · DOI：10.1021/acs.jcim.5c01734 · 被引用次数：2 · 研究领域：Monoclonal and Polyclonal Antibodies Research、vaccines and immunoinformatics approaches、Protein Structure and Dynamics

Quantitative prediction of binding affinity in protein-protein interactions is critical for deciphering biological mechanisms and advancing therapeutic antibody development. While experimental methods for measuring binding affinity remain limited by high-cost, low-throughput constraints, deep learning offers a promising alternative. This study proposes SSIF-affinity, an innovative multimodal deep learning framework that enables high-precision prediction of protein-protein complex binding affinity. First, this method innovatively locates the binding interface and constructs a geometrically constrained binding region, screening key atoms and residues within the binding region. Construct structural diagram data for the selected key atoms to extract atomic-level protein complex interaction features. Second, through the structure-guided cross modal attention module, the structural and sequence features of the selected key residues are fused to capture the structural interactions between residues and the correlations between sequence evolutions. In addition, extracting features of full-length sequence information through convolutional neural network (CNN) and long short-term memory (LSTM) network not only captures local interaction features between sequences, but also mines long-range dependencies through temporal modeling. The final integrated multilevel feature input multilayer perceptron (MLP) regression module predicts the binding affinity values of protein-protein complexes. S...