Improving The Energy Function Used In Rosetta For Protein Structure Prediction and Design

Abstract

Protein structure prediction and design relies on conformational sampling and scoring to uncover a global energy minimum. While it is critical to have an accurate energy function for efficient and precise modeling, the Rosetta energy function trades the accuracy of quantum mechanical modeling for a substantially faster combination of approximated score terms. While the Rosetta energy function has been successfully applied to problems such as ab initio folding, novel fold design, catalytic enzyme design, and numerous other projects, recent work utilizing intensive sampling algorithms have identified instances where native structures are not at the global energy minima. Here, we address two known issues in the energy function: nonideal bond geometry modeling and double counting between the statistical sidechain torsion potential and physical score terms. We also introduce rigorous and sensitive methods designed to quantify energy function performance and uncover errors.