Many drug molecules contain biaryl fragments, resulting in a torsional barrier corresponding to rotation around the bond linking the aryls. The potential energy surfaces of these torsions vary significantly due to steric and electronic effects, ultimately affecting the relative stability of the molecular conformations in the protein-bound and solution states. Simulations of protein-ligand binding require accurate computational models to represent the intramolecular interactions to provide accurate predictions of the structure and dynamics of binding. In this paper, we compare four force fields (Generalized AMBER Force Field (GAFF), Open Force Field (OpenFF), CHARMM General Force Field (CGenFF), Optimized Potentials for Liquid Simulations (OPLS)) and two neural network potentials (ANI-2x, ANI-1ccx) in their ability to predict the torsional potential energy surfaces of 88 biaryls extracted from drug fragments. The mean of the absolute deviation over the full PES (MADF) and the mean absolute deviation of the torsion rotational barrier height (MADB) relative to high-level ab initio reference data was used as a measure of accuracy. In comparison to high-level ab-initio data, ANI-1ccx was most accurate for predicting the barrier height (MADF: 0.5 kcal/mol, MADB: 0.8 kcal/mol), followed closely by ANI-2x (MADF: 0.5 kcal/mol,

Neural Network Potentials (NNPs) have quickly emerged as powerful computational methods for modeling large chemical systems with the accuracy of quantum mechanical methods but at a much smaller computational cost. To make the training and evaluation of the underlying neural networks practical, these methods commonly cutoff interatomic interactions at a modest range (e.g., 5~\AA), so longer-range interactions like London dispersion are neglected. This limits the accuracy of these models for intermolecular interactions. In this work, we develop a new NNP designed for modeling chemical systems were dispersion is an essential component. This new NNP is extended to treat dispersion interactions rigorously by calculating atomic dispersion coefficients through a second NN, which is trained to reproduce the coefficients from the quantum-mechanically derived exchange-hole dipole moment (XDM) model. Calculation of the dispersion component of intermolecular interactions through this scheme provides results in very good agreement with the QM data, with a mean absolute error (MAE) of 0.6 kcal/mol and a coefficient of determination (R2) of 0.98. The dispersion components of these intermolecular interactions are predicted in excellent agreement with the QM data, with a mean absolute error (MAE) of 0.02 kcal/mol and an R2 of 1.00. This combined dispersion-corrected NNP, called ANIPBE0-MLXDM, predicts intermolecular interaction energies for complexes from the DE370K test set with an MAE of 0.5 kcal/mol and an R2 of 0.94 relative to high-level ab initio results (CCSD(T)/CBS), but with a computational cost that is billions of times smaller. The ANIPBE0-MLXDM method is effective for simulating large-scale dispersion-driven systems, like gas adsorption in porous materials, molecular liquids, and nanostructures, on a single computer workstation.

Many drug molecules contain biaryl fragments, resulting in a torsional barrier corresponding to rotation around the bond linking the aryls. The potential energy surfaces of these torsions vary significantly due to steric and electronic effects, ultimately affecting the relative stability of the molecular conformations in the protein-bound and solution states. Simulations of protein--ligand binding require accurate computational models to represent the intramolecular interactions to provide accurate predictions of the structure and dynamics of binding. In this paper, we compare four force fields (Generalized AMBER Force Field (GAFF), Open Force Field (OpenFF), CHARMM General Force Field (CGenFF), Optimized Potentials for Liquid Simulations (OPLS)) and two neural network potentials (ANI-2x, ANI-1ccx) in their ability to predict the torsional potential energy surfaces of 88 biaryls extracted from drug fragments. The mean of the absolute deviation over the full PES (MADF) and the mean absolute deviation of the torsion rotational barrier height (MADB) relative to high-level ab initio reference data was used as a measure of accuracy. In comparison to high-level ab-initio data, ANI-1ccx was most accurate for predicting the barrier height (MADF: 0.5~kcal/mol, MADB:~0.8~kcal/mol), followed closely by ANI-2x (MADF: 0.5~kcal/mol, MADB:~1.0~kcal/mol), then CGenFF (MADF: 0.8~kcal/mol, MADB: 1.3~kcal/mol), OpenFF (MADF: 1.5~kcal/mol, MADB: 1.4~kcal/mol), GAFF (MADF: 1.2~kcal/mol, MADB: 2.6~kcal/mol), and finally OPLS (MADF: 1.5~kcal/mol, MADB: 2.8~kcal/mol). Significantly, the NNPs are systematically more accurate and more reliable than any of the force fields. As a practical example, the neural network potential/molecular mechanics (NNP/MM) method was used to simulate the isomerization of ozanimod, a drug used for multiple sclerosis. Multi-nanosecond molecular dynamics (MD) simulations in an explicit aqueous solvent were performed, as well as umbrella sampling and adaptive biasing force enhanced sampling techniques. These theories predicted a rate of isomerization of $4.30 \times 10^{-1}$ ns$^{-1}$, which is consistent with direct MD simulations.

Neural Network Potentials (NNPs) have quickly emerged as powerful computational methods for modeling large chemical systems with the accuracy of quantum mechanical methods but at a much smaller computational cost. To make the training and evaluation of the underlying neural networks practical, these methods commonly cut off interatomic interactions at a modest range (e.g., 5.2 Å), so longer-range interactions like London dispersion are neglected. This limits the accuracy of these models for intermolecular interactions. In this work, we develop a new NNP designed for modeling chemical systems where dispersion is an essential component. This new NNP is extended to treat dispersion interactions rigorously by calculating atomic dispersion coefficients through a second set of NNs, which is trained to reproduce the coefficients from the quantum-mechanically derived exchange-hole dipole moment (XDM) model. The NNP with this dispersion correction predicts intermolecular interactions in very good agreement with the QM data, with a mean absolute error (MAE) of 0.67 kcal/mol and a coefficient of determination (R2) of 0.97. The dispersion components of these intermolecular interactions are predicted in excellent agreement with the QM data, with a mean absolute error (MAE) of 0.01 kcal/mol and an R2 of 1.00. This combined dispersion-corrected NNP, called ANIPBE0-MLXDM, predicts intermolecular interaction energies for complexes from the DES370K test set with an MAE of 0.69 kcal/mol and an R2 of 0.97 relative to high-level ab initio results (CCSD(T)), but with a computational cost that is billions of times smaller. The ANIPBE0-MLXDM method is effective for simulating large-scale dispersion-driven systems, such as molecular liquids and gas adsorption in porous materials, on a single computer workstation.

Many drug molecules contain biaryl fragments, resulting in a torsional barrier corresponding to rotation around the bond linking the aryls. The potential energy surfaces of these torsions vary significantly due to steric and electronic effects, ultimately affecting the relative stability of the molecular conformations in the protein-bound and solution states. Simulations of protein--ligand binding require accurate computational models to represent the intramolecular interactions to provide accurate predictions of the structure and dynamics of binding. In this paper, we compare four force fields (Generalized AMBER Force Field (GAFF), Open Force Field (OpenFF), CHARMM General Force Field (CGenFF), Optimized Potentials for Liquid Simulations (OPLS)) and two neural network potentials (ANI-2x, ANI-1ccx) in their ability to predict the torsional potential energy surfaces of 88 biaryls extracted from drug fragments. The mean of the absolute deviation over the full PES (MADF) and the mean absolute deviation of the torsion rotational barrier height (MADB) relative to high-level ab initio reference data was used as a measure of accuracy. In comparison to high-level ab-initio data, ANI-1ccx was most accurate for predicting the barrier height (MADF: 0.5~kcal/mol, MADB:~0.8~kcal/mol), followed closely by ANI-2x (MADF: 0.5~kcal/mol, MADB:~1.0~kcal/mol), then CGenFF (MADF: 0.8~kcal/mol, MADB: 1.3~kcal/mol), OpenFF (MADF: 1.5~kcal/mol, MADB: 1.4~kcal/mol), GAFF (MADF: 1.2~kcal/mol, MADB: 2.6~kcal/mol), and finally OPLS (MADF: 1.5~kcal/mol, MADB: 2.8~kcal/mol). Significantly, the NNPs are systematically more accurate and more reliable than any of the force fields. As a practical example, the neural network potential/molecular mechanics (NNP/MM) method was used to simulate the isomerization of ozanimod, a drug used for multiple sclerosis. Multi-nanosecond molecular dynamics (MD) simulations in an explicit aqueous solvent were performed, as well as umbrella sampling and adaptive biasing force enhanced sampling techniques. These theories predicted a rate of isomerization of $4.30 \times 10^{-1}$ ns$^{-1}$, which is consistent with direct MD simulations.