Accurate Deep Learning-Aided Density-Free Strategy for Many-Body Dispersion-Corrected Density Functional Theory

Abstract

Using a deep neuronal network (DNN) model trained on the large ANI-1 data set of small organic molecules, we propose a transferable density-free many-body dispersion (DNN-MBD) model. The DNN strategy bypasses the explicit Hirshfeld partitioning of the Kohn−Sham electron density required by MBD models to obtain the atom-in-molecules volumes used by the Tkatchenko−Scheffler polarizability rescaling. The resulting DNN-MBD model is trained with minimal basis iterative Stockholder atomic volumes and, coupled to density functional theory (DFT), exhibits comparable (if not greater) accuracy to other approaches based on different partitioning schemes. Implemented in the Tinker-HP package, the DNN-MBD model decreases the overall computational cost compared to MBD models where the explicit density partitioning is performed. Its coupling with the recently introduced Stochastic formulation of the MBD equations (J. Chem. Theory Comput. 2022, 18 (3), 1633−1645) enables large routine dispersion-corrected DFT calculations at preserved accuracy. Furthermore, the DNN electron density-free features extend the MBD model's applicability beyond electronic structure theory within methodologies such as force fields and neural networks.Letter pubs.acs.org/JPCL

A Hybrid Machine Learning Approach for Structure Stability Prediction in Molecular Co-crystal Screenings

“…An efficient ML-based MBD implementation that makes this computationally feasible has recently been reported. 55 , 56 …”

“…An efficient MLbased MBD implementation that makes this computationally feasible has recently been reported. 56,57 5. COMPUTATIONAL DETAILS DFT calculations were performed with the all-electron code FHI-aims, 58 using the PBE 45 and PBE0 40 functionals.…”

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

“…It includes direct neural networks coupling with physics-driven contributions going beyond multipolar electrostatics and polarization through inclusion of many-body dispersion models. 71,72 As Deep-HP's purpose is to push a trained ML/hybrid model towards large scale production simulations, we expect extensions of the present simulation capabilities to other class of systems towards materials and catalysis applications. Overall, the present ANI/AMOEBA hybrid model goes a step further towards the grail of molecular mechanics which is the unification within a many-body interaction potential of the short-range quantum mechanical accuracy and of the physically motivated long-range effects at force field cost.…”

Second Quantization Approach to Many-Body Dispersion Interactions

“…The MBD method can be now routinely applied to systems with up to N ∼ 10 4 atoms [9], a size limitation owing to the N 3 computational scaling of MBD. Furthermore, MBD effects have been shown to extrapolate to mesoscale processes [28,29], including solvation and folding of proteins [9,30] and the delamination of graphene from surfaces [31], demonstrating the interplay between MBD modes and collective nuclear vibrations [29,32]. These findings suggest that MBD interactions contribute to cooperative effects between electronic and nuclear degrees of freedom in complex chemical and biophysical systems.…”

“…These effects include nonlocal allosteric pathways in enzymes from coordinated electronic fluctuations [33][34][35] and the emergence of giant electric-dipole oscillations in biomolecules that mediate long-range intermolecular interactions [36,37]. Pursuing the study of MBD effects in realistic systems in complex environments requires simulations with millions of atoms, which are infeasible at the moment even with stochastic implementations [30,38]. The development of a coarse-grained MBD model would be a compelling strategy to provide a conceptual and computational leap to extend the applicability of MBD to million-atom systems.…”