Performance efficient macromolecular mechanics via sub-nanometer shape based coarse graining

Abstract

Dimensionality reduction via coarse grain modeling has positioned itself as an indispensable tool for decades, particularly for biomolecular simulations where atomic systems encompass hundreds of millions of atoms. While distinct flavors of coarse grain modeling exist, those occupying the coarse end of the spectrum are typically knowledge based, relying on a priori information to parameterize models, thus hindering general predictive capability. Here, we present an algorithmic and transferable approach known as shape based coarse graining (SBCG) which employs unsupervised machine learning via competitive Hebbian adaptation to construct coarse molecules that perfectly represent atomistic topologies. We show how SBCG provides ample control over model granularity, and we provide a quantitative metric for selection thereof. Parameter optimization, inclusion of small molecule species, as well as simulation configuration are discussed in detail. Our method and its implementation is made available as part of the CGBuilder plugin, present in the widely-used visual molecular dynamics (VMD) and nanoscale molecular dynamics (NAMD) software suites. We demonstrate applications of our method with a variety of systems from the inositol hexaphosphate-bound, full-scale HIV-1 capsid to heteromultimeric cofilin-2-bound actin filaments. Overall, we show that SBCG provides a simple yet robust approach to coarse graining that requires minimal user input and lacks any ad hoc interactions between protein domains. Furthermore, because the Hamiltonian employed in SBCG is CHARMM compatible, SBCG takes full advantage of the latest GPU-accelerated NAMD3 yielding molecular sampling of over a microsecond per day for systems that span micrometers.