OPUS-Fold3: a gradient-based protein all-atom folding and docking framework on TensorFlow

Abstract

Protein folding and docking framework is crucial for computational structural biology. It can deliver a corresponding 3D structure using given constrains (e.g. distance and orientation distribution constraints proposed by trRosetta). In this paper, we propose OPUS-Fold3, a gradient-based all-atom folding and docking framework. OPUS-Fold3 is capable of modeling protein backbone and side chains either separately or simultaneously using given constrains. As a docking framework, OPUS-Fold3 is also capable of dealing with the constrains between the receptor and ligand. In addition, if a constrain (or potential function) can be represented as a function of heavy atoms' position, it can be easily introduced into OPUS-Fold3 to further improve the folding and docking accuracy. OPUS-Fold3 is written in Python and TensorFlow2.4, which is user-friendly to any source-code level modification, and can be conveniently incorporated with other TensorFlow-based models.