Optimal coarse-grained site selection in elastic network models of biomolecules

TL;DR

This paper introduces a stochastic optimization method for selecting coarse-grained sites in elastic network models of biomolecules, improving model accuracy by systematically choosing sites based on a cost function.

Contribution

The study presents a novel stochastic search strategy for optimal site selection in elastic network models, surpassing traditional intuitive methods.

Findings

Optimized site selection improves model consistency with exact harmonic models.

Algorithm-driven selection significantly impacts biomolecular model properties.

Method applied successfully to both protein and RNA structures.

Abstract

Elastic network models, simple structure-based representations of biomolecules where atoms interact via short-range harmonic potentials, provide great insight into a molecule's internal dynamics and mechanical properties at extremely low computational cost. Their efficiency and effectiveness have made them a pivotal instrument in the computer-aided study of proteins and, since a few years, also of nucleic acids. In general, the coarse-grained sites, i.e. those effective force centres onto which the all-atom structure is mapped, are constructed based on intuitive rules: a typical choice for proteins is to retain only the C$_\alpha$ atoms of each amino acid. However, a mapping strategy relying only on the atom type and not the local properties of its embedding can be suboptimal compared to a more careful selection. Here we present a strategy in which the subset of atoms, each of which is mapped onto a unique coarse-grained site of the model, is selected in a stochastic search aimed at optimising a cost function. The latter is taken to be a simple measure of the consistency between the harmonic approximation of an elastic network model and the harmonic model obtained through exact integration of the discarded degrees of freedom. The method is applied to two representatives of structurally very different types of biomolecules: the protein Adenylate kinase and the RNA molecule adenine riboswitch. Our analysis quantifies the substantial impact that an algorithm-driven selection of coarse-grained sites can have on a model's properties.