Data, models, and additional code for MDtrajNet training and validation

Data, models, and additional code for training and validating MDtrajNet for publication Fuchun Ge, Pavlo O. Dral*. Artificial Intelligence for Direct Prediction of Molecular Dynamics Across Chemical Space. 2025. Preprint on arXiv: https://arxiv.org/abs/2505.16301 (2025.05.22). This repository contains: MD trajectory files propagated with MLatom, in Python npz format. The file names are descriptive. The files contain trajectory snapshots with xyz coordinates, species, and time. They can be loaded as, e.g., trajs = [dict(np.load(f'/mlatom/projects/4D/datasets/trajs/ethanol/traj{i+1}.npz')) for i in range(3)]; model.train(trajectories=trajs,...). The MDtrajNet model files in .pt format. They can be loaded with MLatom (https://github.com/dralgroup/mlatom). The custom code lj.py for generating the reference trajectory with the LJ potential. It uses the custom model class passed to MLatom. The file names are self-descriptive.