NepoIP/MM: Toward Accurate Biomolecular Simulation with a Machine Learning/Molecular Mechanics Model Incorporating Polarization Effects DOI
Ge Song, Weitao Yang

Journal of Chemical Theory and Computation, Год журнала: 2025, Номер unknown

Опубликована: Май 21, 2025

Machine learning force fields offer the ability to simulate biomolecules with quantum mechanical accuracy while significantly reducing computational costs, attracting a growing amount of attention in biophysics. Meanwhile, by leveraging efficiency molecular mechanics modeling solvent molecules and long-range interactions, hybrid machine learning/molecular (ML/MM) model offers more realistic approach describing complex biomolecular systems solution. However, multiscale models electrostatic embedding require accounting for polarization ML region induced MM environment. To address this, we adapt state-of-the-art NequIP architecture into polarizable field, NepoIP, enabling effects based on external potential. We found that nanosecond MD simulations NepoIP/MM are stable periodic solvated dipeptide system, converged sampling shows excellent agreement reference QM/MM level. Moreover, show single NepoIP can be transferable across different fields, as well an extremely environment water proteins, laying foundation developing general field used ML/MM embedding.

Язык: Английский

NepoIP/MM: Toward Accurate Biomolecular Simulation with a Machine Learning/Molecular Mechanics Model Incorporating Polarization Effects DOI
Ge Song, Weitao Yang

Journal of Chemical Theory and Computation, Год журнала: 2025, Номер unknown

Опубликована: Май 21, 2025

Machine learning force fields offer the ability to simulate biomolecules with quantum mechanical accuracy while significantly reducing computational costs, attracting a growing amount of attention in biophysics. Meanwhile, by leveraging efficiency molecular mechanics modeling solvent molecules and long-range interactions, hybrid machine learning/molecular (ML/MM) model offers more realistic approach describing complex biomolecular systems solution. However, multiscale models electrostatic embedding require accounting for polarization ML region induced MM environment. To address this, we adapt state-of-the-art NequIP architecture into polarizable field, NepoIP, enabling effects based on external potential. We found that nanosecond MD simulations NepoIP/MM are stable periodic solvated dipeptide system, converged sampling shows excellent agreement reference QM/MM level. Moreover, show single NepoIP can be transferable across different fields, as well an extremely environment water proteins, laying foundation developing general field used ML/MM embedding.

Язык: Английский

Процитировано

0