xTrimoDock: Cross-Modal Transformer for Multi-Chain Protein Docking

Ruijia Wang, Shaochuan Li, Xiao Wang, Zhiyuan Lu, Bing Yang, Hui Li, Chuan Shi, Le Song

Abstract: The structure of a protein–protein complex plays a critical role in understanding the dynamics of binding, delineating biological mechanisms, and developing intervention strategies. Rigid protein-protein docking, assuming no conformational change within proteins, predicts the 3D structure of protein complexes from unbound chains. According to the number of chains, rigid docking is divided into binary complex setting that contains only two chains, and more ubiquitous multi-chain complex setting. Most existing docking methods are tailored for binary complexes, and are computationally expensive or not guaranteed to find accurate complex structures. In this paper, we propose a novel model xTrimoDock for the docking of multi-chain complexes, which can simultaneously employ information from both sequence modality and structure modality of involved protein chains. Specifically, xTrimoDock leverages a cross-modal transformer to integrate representations from protein sequences and structures, and conducts a multi-step prediction of rotations and translations to accomplish the multi-chain docking. Extensive experiments reflect the promising results of the proposed model in the harder multi-chain complex setting.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics

Submission Guidelines: Yes

Please Choose The Closest Area That Your Submission Falls Into: Machine Learning for Sciences (eg biology, physics, health sciences, social sciences, climate/sustainability )