Scalable Hybrid Deep Neural Networks/Polarizable Potentials Biomolecular Simulations including long-range effects

Authors:

Abstract
Deep-HP is a scalable extension of the Tinker-HP multi-GPUs molecular dynamics (MD) package enabling the use of Deep Neural Networks (DNNs). Deep-HP increases DNNs MD capabilities by orders of magnitude offering access to ns simulations for 100k-atom biosystems. It allows to introduce the ANI-2X/AMOEBA hybrid polarizable potential designed for ligand binding studies where solvent-solvent/solvent-solute interactions are computed with the AMOEBA polarizable force field while solute-solute ones are computed by the ANI-2x DNN. ANI-2X/AMOEBA explicitly includes AMOEBA's physical long-range interactions while preserving ANI-2X's solute short-range quantum mechanical accuracy. ANI-2X/AMOEBA is accelerated using multiple-time-stepping, primarily evaluating AMOEBA forces while including ANI-2X ones via correction steps. Simulating more than 10 $\mu$s, we compute charges/uncharged ligands solvation free energies in 4 solvents, and absolute binding free energies of host-guest complexes from SAMPL challenges. ANI-2X/AMOEBA average errors are within chemical accuracy opening the path towards large-scale hybrid DNNs simulations, at force field cost, in biophysics/drug discovery.