Protein structure prediction by an iterative search method

TL;DR

This paper introduces the difference map algorithm for protein structure prediction, which efficiently finds low-energy conformations by intersecting peptide geometry and energy constraints, outperforming existing methods.

Contribution

The paper presents a novel iterative search method, the difference map, for protein conformation prediction that improves the rate of finding low-energy structures.

Findings

Difference map finds low-energy conformations faster than parallel tempering.

The algorithm effectively intersects geometric and energetic constraints.

Results show significant improvement over existing search algorithms.

Abstract

We demonstrate a new algorithm for finding protein conformations that minimize a non-bonded energy function. The new algorithm, called the difference map, seeks to find an atomic configuration that is simultaneously in two constraint spaces. The first constraint space is the space of atomic configurations that have a valid peptide geometry, while the second is the space of configurations that have a non-bonded energy below a given target. These two constraint spaces are used to define a deterministic dynamical system, whose fixed points produce atomic configurations in the intersection of the two constraint spaces. The rate at which the difference map produces low energy protein conformations is compared with that of a contemporary search algorithm, parallel tempering. The results indicate the difference map finds low energy protein conformations at a significantly higher rate then parallel tempering.