Data for Teachers that teach the irrelevant: Pre-training machine learned interaction potentials with classical force fields for robust molecular dynamics simulations

This repository contains datasets, trained models, simulation scripts, and analysis notebooks for molecular dynamics (MD), metadynamics, and machine learning potential development using NewtonNet. The data spans three chemical systems—aspirin, water, and H₂ combustion reactions—and includes training, sampling, and simulation workflows.
Key contents:
Rattled sampling with GAFF/Q-force: Structure files and scripts for generating diverse molecular configurations of aspirin, water, and H₂ combustion systems.Trained NewtonNet models: Collections of models trained from scratch and via pre-training/fine-tuning strategies on datasets such as MD17, FANE, GAFF, Q-force, and H2comb data.Molecular dynamics simulations: Python scripts and Jupyter notebooks for running and analyzing MD simulations with NewtonNet-trained models on aspirin and water.Metadynamics simulations: Input files, scripts, and trajectories for PLUMED-driven metadynamics on H2 combustion reactions, along with analysis notebooks.Simulation trajectories and outputs: MD and metadynamics trajectory data generated from trained models across all systems.This dataset supports benchmarking of machine learning interatomic potentials with and without transfer learning, and provides workflows for molecular sampling, training, simulation, and enhanced sampling across diverse chemical systems.