Teaching computers to fold proteins

TL;DR

This paper presents a gradient-based optimization algorithm for potential functions in protein folding, improving the accuracy of predicted native structures using Monte Carlo estimates on a Lennard-Jones force field.

Contribution

It introduces a novel thermodynamic gradient optimization method for refining potential functions in protein folding simulations.

Findings

Two-thirds of tested peptides folded within 3Å of native structure after optimization.

Initial potential functions failed to fold peptides correctly.

The method effectively improves folding predictions through iterative learning.

Abstract

A new general algorithm for optimization of potential functions for protein folding is introduced. It is based upon gradient optimization of the thermodynamic stability of native folds of a training set of proteins with known structure. The iterative update rule contains two thermodynamic averages which are estimated by (generalized ensemble) Monte Carlo. We test the learning algorithm on a Lennard-Jones (LJ) force field with a torsional angle degrees-of-freedom and a single-atom side-chain. In a test with 24 peptides of known structure, none folded correctly with the initial potential functions, but two-thirds came within 3{\AA} to their native fold after optimizing the potential functions.