QM_eMMa: An Open GPU-Accelerated Framework for Ab Initio QM/MM Molecular Dynamics and Enhanced Sampling
- 1Computing for Science
- 2Institut Laue-Langevin
- 3Otto-Loewi Research Center, Medical University of Graz, Neue Stiftingtalstrasse 6/III, Graz, Austria, A-8010
Abstract
Ab initio QM/MM molecular dynamics can describe chemical processes accurately, but its practical use is limited by the cost of repeated electronic-structure calculations and by overhead in coupling molecular-dynamics and quantum-chemistry programs. We present QM_eMMa, an open interface that couples AMBER to PySCF/GPU4PySCF while keeping the electronic-structure engine and GPU context active throughout a simulation. The converged electronic density is reused between consecutive MD steps, with automatic recovery through a cold-start SCF calculation when needed. We first optimized the workflow with a GH125 α-mannosidase model. Removing repeated interface overhead increased throughput by approximately 26%. Systematic energy and force comparisons then identified a less expensive numerical setting that retained the accuracy required for MD and increased throughput by a further 23% in independent GH125 simulations. The same protocol enabled stable DFT/MM dynamics of a larger enzymatic QM region containing 115 electronic centers and sustained 145 ps of ab initio well-tempered metadynamics, recovering the principal α-mannose conformational basins observed previously. Finally, a reactive aqueous SN2 benchmark showed that the optimized and stricter protocols gave activation and reaction free energies differing by only 0.73 and 0.01 kcal mol-1, respectively, while the optimized calculation remained faster. QM_eMMa therefore provides a practical open framework for single-GPU DFT/MM dynamics and enhanced sampling, with numerical approximations selected through explicit validation of energies, forces, dynamics, and chemical reactivity.
Related
- OpenFerro: a unified framework for mesoscale ferroic dynamics and enhanced sampling
- Accelerated "on-the-fly" coupled-cluster path-integral molecular dynamics: Impact of nuclear quantum effects on an asymmetric proton
- GPU-Accelerated Large-Scale Nonadiabatic Excited State Molecular Dynamics
- LAO+: Focused, Multi-Replica Accelerated Orthogonal Sampling for Robust Binding Free Energy Predictions
- Electrochemical Molecular Dynamics within the Finite-Field Framework
BibTeX
@misc{alonsogil2026qmemma,
title = {{QM_eMMa: An Open GPU-Accelerated Framework for Ab Initio QM/MM Molecular Dynamics and Enhanced Sampling}},
author = {Santiago Alonso-Gil},
howpublished = {ChemRxiv},
year = {2026},
month = sep,
doi = {10.26434/chemrxiv.15009206/v1},
url = {https://chemrxiv.org/doi/full/10.26434/chemrxiv.15009206/v1},
note = {CC BY-NC-ND},
}
RIS
TY - UNPB
TI - QM_eMMa: An Open GPU-Accelerated Framework for Ab Initio QM/MM Molecular Dynamics and Enhanced Sampling
AU - Santiago Alonso-Gil
PY - 2026
DA - 2026/09/21
PB - ChemRxiv
DO - 10.26434/chemrxiv.15009206/v1
UR - https://chemrxiv.org/doi/full/10.26434/chemrxiv.15009206/v1
AB - Ab initio QM/MM molecular dynamics can describe chemical processes accurately, but its practical use is limited by the cost of repeated electronic-structure calculations and by overhead in coupling molecular-dynamics and quantum-chemistry programs. We present QM_eMMa, an open interface that couples AMBER to PySCF/GPU4PySCF while keeping the electronic-structure engine and GPU context active throughout a simulation. The converged electronic density is reused between consecutive MD steps, with automatic recovery through a cold-start SCF calculation when needed. We first optimized the workflow with a GH125 α-mannosidase model. Removing repeated interface overhead increased throughput by approximately 26%. Systematic energy and force comparisons then identified a less expensive numerical setting that retained the accuracy required for MD and increased throughput by a further 23% in independent GH125 simulations. The same protocol enabled stable DFT/MM dynamics of a larger enzymatic QM region containing 115 electronic centers and sustained 145 ps of ab initio well-tempered metadynamics, recovering the principal α-mannose conformational basins observed previously. Finally, a reactive aqueous SN2 benchmark showed that the optimized and stricter protocols gave activation and reaction free energies differing by only 0.73 and 0.01 kcal mol-1, respectively, while the optimized calculation remained faster. QM_eMMa therefore provides a practical open framework for single-GPU DFT/MM dynamics and enhanced sampling, with numerical approximations selected through explicit validation of energies, forces, dynamics, and chemical reactivity.
ER -