Predicting Three-Dimensional Atomic-Level Structures of Biomolecular Complexes

Abstract

OpenComplex1 is an open-source platform for comprehending and developing biomolecular models. Based on the architectures of AlphaFold2 and AlphaFold-Multimer, the current version of OpenComplex realizes end-to-end predictions of the biomolecular three-dimensional (3D) structures including protein, RNA, and protein-RNA complexes. OpenComplex makes it able to infer and train models of biomolecular complexes in a unified framework and supports encoders that use both MSAs and protein language models (pLM). Through extensive training on both structure and sequence databases such as the Protein Data Bank (PDB) and RNA families (Rfam) database, OpenComplex is able to generate models with equivalent or better accuracy for a wide range of biomolecules including monomers or multimers of protein and RNA as well as protein-RNA complexes. In summary, OpenComplex is a powerful tool for predicting accurate biomolecular structures in diverse application scenarios.