AP-Net: An Atomic-Pairwise Neural Network for Smooth and Transferable Interaction Potentials

Abstract

<div> <div> <div> <p>Intermolecular interactions are critical to many chemical phenomena, but their accurate computation using <i>ab initio</i> methods is often limited by computational cost. The recent emergence of machine learning (ML) potentials may be a promising alternative. Useful ML models should not only estimate accurate interaction energies, but also predict smooth and asymptotically correct potential energy surfaces. However, existing ML models are not guaranteed to obey these constraints. Indeed, systemic deficiencies are apparent in the predictions of our previous hydrogen-bond model as well as the popular ANI-1X model, which we attribute to the use of an atomic energy partition. As a solution, we propose an alternative atomic-pairwise framework specifically for intermolecular ML potentials, and we introduce AP-Net—a neural network model for interaction energies. The AP-Net model is developed using this physically motivated atomic-pairwise paradigm and also exploits the interpretability of symmetry adapted perturbation theory (SAPT). We show that in contrast to other models, AP-Net produces smooth, physically meaningful intermolecular potentials exhibiting correct asymptotic behavior. Initially trained on only a limited number of mostly hydrogen-bonded dimers, AP-Net makes accurate predictions across the chemically diverse S66x8 dataset, demonstrating significant transferability. On a test set including experimental hydrogen-bonded dimers, AP-Net predicts total interaction energies with a mean absolute error of 0.37 kcal mol−1, reducing errors by a factor of 2-5 across SAPT components from previous neural network potentials. The pairwise interaction energies of the model are physically interpretable, and an investigation of predicted electrostatic energies suggests that the model ‘learns’ the physics of hydrogen-bonded interactions. </p> </div> </div> </div>

Gaussian Process Regression for Materials and Molecules

“…Since this is now routine for the ML modeling of isolated molecules (see, e.g., ref ( 180 )), even considering excited states, 347 − 349 it is natural to use the techniques illustrated here to carry over this high level of accuracy to periodic systems. Reference electronic-structure methods and training databases have to be chosen carefully, but it is now within reach to train intermolecular potentials using symmetry adapted perturbation theory, 350 random phase approximation, 164 or even quantum Monte Carlo 351 data.…”

Directional ΔG Neural Network (DrΔG-Net): A Modular Neural Network Approach to Binding Free Energy Prediction

“…Therefore, we expect the same architectural choice will help in the adjacent free energy problem. The software for the current project relies upon some of the infrastructure developed for our most recent version of AP-Net. , Empirically, ligand-based modeling is impressively performant, so it would be useful to adapt ligand-based frameworks to the Cartesian data. Quantities like strain and entropy are not included elsewhere and the signal relating to those contributions is partly contained in the ligand geometry. …”

Machine learning meets chemical physics

“…Glick et al use a pairwise neural network to achieve high accuracy in the description of intermolecular terms, 15 while Metcalf et al tackle directly the problem of predicting interaction energies by learning terms computed by symmetry-adapted perturbation theory decomposition. 16 In Ref. 17, Sauceda et al compare gradient-domain machine learning with conventional force fields to achieve a more efficient implementation of molecular PES.…”