AIMNet-NSE test datasets

Ions-16 and ChEMBL-20 datasets used to test the performance of the AIMNet-NSE model. The datasets contain PBE0/ma-Def2-SVP energies and NBO atomic charges for the non-equilibrium conformers of neutral organic molecules randomly sampled from the PubChem database (Ions-16) and B97-3c optimized conformations of neutral organic molecules randomly sampled from the ChEMBL database (ChEMBL-20). The number in the dataset name corresponds to the maximum number of non-hydrogen atoms in the molecules. The dataset which was used for training the AIMNet-NSE model contains molecules up to 12 non-H atoms, whereas Ions-16 and ChEMBL-20 contain the molecules with 13 non-hydrogen atoms or more.

The datasets are formatted as HDF5 files. Data group names have the format as _???, where ??? corresponds to the number of atoms in molecules. Each group contains data for M molecules, each with N atoms. The groups contain the following datasets:

Name    Data type    Shape      Description
mol_id     S24             M            Molecule ID
coord      float32        M, N, 3    Cartesian coordinates, Å
numbers uint8           M, N         Atomic numbers
charge    int8             M              Molecular charge
mult        unit8           M              Spin multiplicity
energy    float64        M              PBE0/ma-def2-SVP energy, eV
charges   float32       M, N, 2       α and β NBO charges