Machine Learning of coarse-grained Molecular Dynamics Force Fields

• DOI: 10.48550/arxiv.1812.01736
• Publication Date: 2018-01-01
• Authors: Jiang Wang; Simon Olsson; Christoph Wehmeyer; Adrià Pérez; Nicholas E. Charron; Gianni De Fabritiis; Frank Noé; Cecilia Clementi
• Keywords: Granularity

Atomistic or ab-initio molecular dynamics simulations are widely used to predict thermodynamics and kinetics relate them structure. A common approach go beyond the time- length-scales accessible with such computationally expensive is definition of coarse-grained models. Existing coarse-graining approaches define an effective interaction potential match defined properties high-resolution models experimental data. In this paper, we reformulate as a supervised machine learning problem. We use statistical theory decompose error cross-validation select compare performance different introduce CGnets, deep approach, that learns free energy functions can be trained by force matching scheme. CGnets maintain all physically relevant invariances allow one incorporate prior physics knowledge avoid sampling unphysical structures. show capture all-atom explicit-solvent surfaces using only few beads no solvent, while classical methods fail crucial features surface. Thus, able multi-body terms emerge from dimensionality reduction.