Jadwal Sholat

Memuat jadwal sholat…

Ilmu Komputer & AI editorial

Open AccessOA2023

OpenMM 8: Molecular Dynamics Simulation with Machine Learning Potentials

OpenMM 8 introduces support for arbitrary PyTorch machine learning potentials, a high-level interface for pretrained models, and optimized CUDA kernels, enabling accurate simulations of CDK8 and GFP chromophore at modest cost.
Peter K. Eastman; Raimondas Galvelis; Raúl P. Peláez; C. Abreu; Stephen E. Farr; Emilio Gallicchio; Anton Gorenko; Mike Henry; Frank Hu; Jing Huang; Andreas Krämer; Julien Michel; Joshua A. Mitchell; Vijay S. Pande; J. P. Rodrigues; Jaime Rodríguez-Guerra; Andrew C. Simmonett; Sukrit Singh; Jason M. Swails; Philip Turner; Yuan-Qing Wang; Ivy Zhang; J. Chodera; G. de Fabritiis; T. Markland· Journal of Physical Chemistry B· 2023· DOI 10.1021/acs.jpcb.3c06662

The core problem

Machine learning (ML) is increasingly important in molecular simulation, offering the potential to improve accuracy while maintaining efficiency. The OpenMM molecular dynamics toolkit, widely used for biomolecular simulations, has released version 8 with new features specifically designed to support ML potentials. This work presents these features and demonstrates their practicality on two test systems: cyclin-dependent kinase 8 (CDK8) and the green fluorescent protein (GFP) chromophore in water. The motivation is to make ML-based simulations accessible and efficient for a broad range of users.

Innovation

OpenMM 8 allows arbitrary PyTorch models to be incorporated into simulations, computing forces and energy directly from the model. A higher-level interface simplifies the use of general-purpose, pretrained potential functions, enabling users to model molecules of interest without deep ML expertise. To achieve high performance, the release includes a collection of optimized CUDA kernels and custom PyTorch operations. These optimizations reduce the computational overhead typically associated with ML potentials, making simulations practical at only a modest increase in cost compared to traditional force fields.
Introduction
Machine learning (ML) is increasingly important in molecular simulation, offering the potential to improve accuracy while maintaining efficiency. The OpenMM molecular dynamics toolkit, widely used for biomolecular simulations, has released version 8 with new features specifically designed to support ML potentials. This work presents these features and demonstrates their practicality on two test systems: cyclin-dependent kinase 8 (CDK8) and the green fluorescent protein (GFP) chromophore in water. The motivation is to make ML-based simulations accessible and efficient for a broad range of users.

Why it matters

The integration of ML potentials into OpenMM 8 represents a significant step toward routine use of ML in molecular dynamics. By supporting arbitrary PyTorch models and providing a user-friendly interface for pretrained potentials, the toolkit lowers the barrier for researchers to adopt ML-based simulations. The performance optimizations ensure that the increased accuracy from ML potentials comes at a manageable cost. This work enables future applications in drug discovery, protein engineering, and other areas where accurate molecular modeling is critical. The authors conclude that these features make it practical to use machine learning to improve simulation accuracy at modest additional cost.

Who should read this

Pembaca Research Digest dan praktisi rumpun ini.

Opening member content…