Machine learning-accelerated quantum mechanics-based atomistic simulations for industrial applications

Abstract

Atomistic simulations have become an invaluable tool for industrial applications ranging from the optimization of protein-ligand interactions for drug discovery to the design of new materials for energy applications. Here we review recent advances in the use of machine learning (ML) methods for accelerated simulations based on a quantum mechanical (QM) description of the system. We show how recent progress in ML methods has dramatically extended the applicability range of conventional QM-based simulations, allowing to calculate industrially relevant properties with enhanced accuracy, at reduced computational cost, and for length and time scales that would have otherwise not been accessible. We illustrate the benefits of ML-accelerated atomistic simulations for industrial R&D processes by showcasing relevant applications from two very different areas, drug discovery (pharmaceuticals) and energy materials. Writing from the perspective of both a molecular and a materials modeling scientist, this review aims to provide a unified picture of the impact of ML-accelerated atomistic simulations on the pharmaceutical, chemical, and materials industries and gives an outlook on the exciting opportunities that could emerge in the future.

Structure-based in silico approaches for drug discovery against Mycobacterium tuberculosis

“…The main drawbacks of MD simulations are the computational resource demands, the large timescales required to sample many protein conformations and errors due to the underlying force fields. To combat these issues, machine learning has been applied to several different aspects of MD simulations, including improvements to the underlying force fields [113] , [114] , [115] ; increasing the protein conformations which are sampled [116] and improving the analysis of MD simulations [116] , [117] . The application to the improvement of underlying forcefield, while able to achieve large timescale reductions [115] , have so far only been applied to simple organic molecules [113] , [115] , [118] or large single-component systems [119] , rather than biomolecules.…”

“…To combat these issues, machine learning has been applied to several different aspects of MD simulations, including improvements to the underlying force fields [113] , [114] , [115] ; increasing the protein conformations which are sampled [116] and improving the analysis of MD simulations [116] , [117] . The application to the improvement of underlying forcefield, while able to achieve large timescale reductions [115] , have so far only been applied to simple organic molecules [113] , [115] , [118] or large single-component systems [119] , rather than biomolecules. Selected examples for machine learning-based enhanced sampling of protein conformations will be highlighted, as they have been applied most extensively to biomolecular MD simulations.…”

Primary radiation damage in silicon from the viewpoint of a machine learning interatomic potential

“…Hence, the simulation capability that has been made available by ML potentials makes larger scale molecular dynamics simulations with DFT accuracy reachable [30][31][32][33][34][35][36][37][38][39]. That is why they have received a great amount of interest in the community and their implementation in different fields matures rapidly [40][41][42][43][44].…”

AENET-LAMMPS and AENET-TINKER: Interfaces for Accurate and Efficient Molecular Dynamics Simulations with Machine Learning Potentials

“…Due to their ability to include realistic experimental conditions, such as solvent effects, temperature, and pressure, they are routinely employed in a wide range of research areas in both academic and industrial settings. Prominent examples are the identification of novel materials for energy applications [3][4][5][6][7] or the design of novel drugs [8][9][10][11] . To perform these simulations, reliable interatomic potentials (or force fields) are required that describe the atomic interactions 12,13 .…”

“…They need to be accurate to capture the atomistic process of interest but also efficient to generate simulations on sufficient time and length scales using reasonable compute resources. Machine-learning potentials (MLPs) [14][15][16][17][18][19][20][21][22][23][24][25][26] , including approaches based on artificial neural networks (ANNs), learn the interatomic interactions from accurate quantum mechanical (QM) calculations such as density-functional theory (DFT) 27 and have shown great promise in combining accuracy and affordability allowing to simulate complex systems under realistic conditions 11,16,[28][29][30] .…”