Computational Protein Redesign and Decoy Discrimination

A key aim of computational protein design is to understand how amino acid mutations affect the structure and stability of proteins. Recent studies have used molecular dynamics (MD) simulations to predict the response of wildtype proteins to mutations. However, MD structures can be trapped in local free energy basins, resulting in configurations that diverge significantly from experimental crystal structures. In this work, we construct a mutation dataset, which contains 32 pairs of single-core-residue mutated crystal structures and their corresponding wildtype structures. We perform replica-exchange MD (REMD) simulations on wildtype and mutant structures to obtain a series of possible protein conformations before and after the core mutations. We then perform residue repacking to determine whether the side chain conformations match these in the crystal structures. We also evaluate the local packing fraction and void geometry of the protein structures from REMD simulations to distinguish decoys from the crystal structures. The ability to distinguish experimental structures from decoys will enable unprecedented validation of MD simulations, and provide insight into novel structures that have not yet been experimentally characterized.

NIH Training Grant, Grant No.1T32EB019941.