Reconstructing atomistic structures from residue-level protein structure networks using artificial neural networks

Abstract

It is well-established that one of the fundamental challenges in carrying out molecular simulations of biological systems is that the experimentally relevant timescales and system sizes are typically many orders of magnitude larger than what is computationally tractable. One strategy for addressing this challenge is to build coarse-grained representations of the systems of interest, whereby computational speedup is gained by lumping together some of the degrees of freedom (e.g. all the atoms in each amino acid of a protein are unified as a single particle).