Unsupervised Refinement of Protein Structures

Abstract

Protein structure–function relationships are of key interest in biochemistry and biophysics. The determination of structures is a crucial step towards establishing how protein folds and conformations drive the execution of fundamental biological processes. However, atomistic structure determination presents challenges, particularly for large or complex systems. The resolution of experimentally-determined structures impacts the ability of researchers to interpret data and elucidate key details of structure–function relationships. Here, we describe unsupervised approaches for protein structure refinement based on the Rosetta software suite and experimental data obtained by Cryo-EM or NMR, and demonstrate their application in example use cases. Results indicate that these approaches successfully address geometrical outliers in the initial models and significantly improve their structural quality while maintaining consistency with the experimental data. Since the refinement procedure is automated with minimal user input, the approaches are suitable as a pipeline for high-throughput protein structure refinement.