Feshdock: A Divide-and-Conquer Protein–Protein Complex Conformation Prediction Algorithm

Abstract

Understanding protein interactions is vital for unraveling cellular biological processes, disease mechanisms, and drug development. Numerous in-silico protein–protein docking algorithms have been developed to substitute time-consuming and expensive experimental methods for deciphering such interactions. However, few algorithms can provide satisfactory complex conformation prediction performance due to the vast and multimodal landscapes of protein surfaces and to the complicated conformation changes of side chains. This motivated the development of a novel protein–protein docking algorithm called Feshdock, which is introduced in this paper. Specifically, Feshdock leverages the Fast Fourier transform to globally scan potential binding positions, which are then partitioned into multiple clusters using the K-means method. These clusters are locally optimized by a swarm intelligence algorithm in a divide-and-conquer manner, accompanied by an anisotropic network model considering protein flexibility. Finally, hierarchical clustering is employed to determine the final candidate complex conformations. The effectiveness of our proposed Feshdock approach is verified on the benchmark dataset of 230 protein complexes. Our experimental results show that the proposed approach performs better than almost all commonly used docking methods containing ZDOCK, MEGADOCK, and LightDock. In addition, being a lightweight algorithm, Feshdock is still very competitive in terms of docking accuracy and efficiency, being even comparable to emergent deep learning-based approaches, including ColabFold and AlphaFold2-refined ClusPro. The source code of Feshdock can be downloaded for free at https://github.com/SZU-ADDG/Feshdock.