Simulating protein–ligand binding with neural network potentials

Abstract

Neural network potentials provide accurate predictions of the structures and stabilities of drug molecules. We present a method to use these new potentials in simulations of drugs binding to proteins using existing molecular simulation codes.

Profiling SARS-CoV-2 Main Protease (MPRO) Binding to Repurposed Drugs Using Molecular Dynamics Simulations in Classical and Neural Network-Trained Force Fields

“…We combined the accurate ANI-2x force field for drugs with the CHARMM36m/TIP3P force fields for proteins and solvent to run hybrid ANI/MM MD simulations 31 of the M PRO -drug complexes ( Figure 2 ), as implemented in the NAMD package. 32 In these hybrid ANI/MM simulations, the total potential energy ( U ) of the system was defined as the sum of the energies of the ANI region (i.e., the drug molecule) and the MM region (protein and solvent) and the interaction energy between the drug and the MM region: 31 …”

“…The U ANI/MM ( r ANI , r MM ) term comprised MM nonbonded interactions between the MM region and the drug, that is, Coulombic and Lennard-Jones interactions between the ANI atoms and MM atoms: 31 …”

Towards chemical accuracy for alchemical free energy calculations with hybrid physics-based machine learning / molecular mechanics potentials

“…the MM/ML potential energy function for the ligand and U MM (X , X ) the MM potential energy function for the environment/ligand system. The formulation given in Equation 1 treats intramolecular ligand interactions with an ML potential while intermolecular and environmental (receptor and/or solvent) atomic interactions are treated with MM force fields[52]. Although other formulations are possible-such as including short-range ligand-environment interactions within the ML region-we will demonstrate that the simple formulation of Equation 1 is sufficient to realize significant improvements in the accuracy of computed binding free energies for a challenging kinase:inhibitor benchmark system (Figure 2).Nonequilibrium perturbations can efficiently compute MM to ML/MM correctionsCurrent implementations of ML potentials do not permit an entire alchemical free energy calculation to be carried out with hybrid ML/MM potentials in a practical timescale.…”

A machine learning based intramolecular potential for a flexible organic molecule

“…43,44 So far these studies have tended to focus on gas phase dynamics, however a recent study showed that neural-network potentials (within a QM/MM type approach) are capable of predicting the structural conformations of drugs in protein binding pockets, as well as conformational components of binding free energies. 45 Interestingly, this study showed that conformational binding energies can be over-estimated by molecular mechanics force elds by several kcal mol À1 . But otherwise, relatively little is known about the performance of these machine learning based potentials in the condensed phase, where free energy basins that are unpopulated in the gas phase may emerge.…”

“…where E L represents the intramolecular energy of the ligand, E R is the potential energy of the receptor, including water molecules, and E RL is the interaction energy between the ligand and the receptor. Similar to a hybrid QM/MM simulation set-up (and also the approach taken recently with a neural network potential 45 ), we treat the various energetic components using different levels of theory. The protein environment is described using the OPLS-AA/M force eld, 14 and water molecules are described using the TIP4P model.…”