GEOM-QM9 pre-simulated MD trajectories (Align Your Structures, ICLR 2026)

Pre-simulated molecular-dynamics trajectories for the GEOM-QM9 subset used in Align Your Structures: Generating Trajectories with Structure Pretraining for Molecular Dynamics (ICLR 2026). Contents: 11,807 per-molecule simulation directories (5,534 train / 5,080 val / 1,193 test), corresponding to the 1,109 / 1,018 / 239 unique SMILES described in §5.2 of the paper. Each directory contains mol.pkl (RDKit Mol with hydrogens), system.pdb (OpenMM topology + initial positions), traj.xtc (5 ns trajectory at 4 fs timestep, frame_interval=100), and smiles.txt. Three *_mol_dirs.pkl split files under splits/ allow deterministic loader reproduction. Simulation settings: all-atom, explicit-solvent, 5 ns per trajectory, 4 fs timestep, hydrogen mass repartitioning at 1.5 amu, 100-step frame interval. Five independent replicas per SMILES. Reference code: https://github.com/ani11452/Align_Your_Structures If you use this data please cite the paper above and the upstream GEOM dataset (Axelrod & Gómez-Bombarelli, Scientific Data 2022, doi:10.7910/DVN/JNGTDF).