Speaker
Description
Hybrid machine learning/molecular mechanical (ML/MM) simulations are increasingly explored as an alternative to computationally expensive quantum mechanical (QM) methods, offering near first-principles accuracy at substantially reduced computational cost. Neural network potentials (NNPs) can reproduce QM-level energies and forces efficiently, yet their integration into established molecular dynamics (MD) engines has often been limited. To address this, we here present an NNP-interface implemented in the widely used MD code GROMACS. The interface enables NNPs trained in the PyTorch framework to contribute energies and forces during MD simulations, either for selected subsets or entire molecular systems. In particular, the design integrates NNP inference seamlessly into the extensive GROMACS molecular simulation ecosystem, providing users with the capability to straightforwardly combine NNPs with existing advanced sampling and free energy workflows.
In this talk, we will review the basic concepts of neural network potentials, and demonstrate the capabilities of our interface using several representative applications, including performance benchmarks, enhanced sampling of peptide torsional free energy landscapes, absolute solvation free energy calculations, and protein-ligand simulations.
Lastly, we will present ongoing work on a large membrane protein in the mammalian brain, illustrating how NNPs are expanding the possibilities of studying complex biomolecular processes.