Supporting Datasets for: Stable and Accurate Atomistic Simulations of Flexible Molecules using Conformationally Generalisable Machine Learned Potentials

Supporting datasets for: Stable and Accurate Atomistic Simulations of Flexible Molecules using Conformationally Generalisable Machine Learned Potentials.For each of aspirin, paracetamol and salicylic acid there are 5 datasets. Three of the datasets (MD-300K, MD-500K and Meta-300K) contain 10,000 structures; these were generated by sampling every 1 ps from 10 ns molecular dynamics simulations in the gas phase using GAFF. Two of the datasets (phi-scan and psi-scan) contain 72 structures; these were generated by rotating the dihedral angles defined by φ and ψ in the paper. All energies and forces were calculated at the B3LYP-D3/6-31G* level of theory. For each dataset, energies in energies.txt are given in units of kcal/mol, Cartesian atomic forces in forces.txt are given in units of kcal/mol/ångstrom and coordinates in coords.txt are given in units of ångstrom. Nuclear charges, which define the atom ordering, are provided in nuclear_charges.txt.