xnn.common.data.hub.ani2x#

ANI-2x dataset builder (pyanitools HDF5 from the Zenodo release).

The ANI-2x data set (Devereux et al., J. Chem. Theory Comput. 16, 4192, 2020) extends ANI to seven elements (adds S, F, Cl to H, C, N, O). It holds ~9.6 million active-learning-selected off-equilibrium conformations with wB97X/6-31G(d) energies and forces, the level the ANI-2x model (xnn.dnn.models.ani.ANI.ani2x()) was trained on. Upstream distributes it as a ~3.7 GB ANI-2x-wB97X-631Gd.tar.gz archive holding one pyanitools HDF5 file whose top-level groups are keyed by number of atoms ("002", "003", …). Each group holds coordinates (Nc, Na, 3), species (Nc, Na) (as atomic numbers), energies (Nc,) and forces (Nc, Na, 3) – so within a group each conformation can be a different molecule with the same atom count.

Reference#

Devereux et al., “Extending the Applicability of the ANI Deep Learning Molecular Potential to Sulfur and Halogens”, J. Chem. Theory Comput. 16, 4192 (2020). Data: https://doi.org/10.5281/zenodo.10108942 Format / reader spec: the Zenodo record’s sample_data_loader.py and supplementary_information.pdf (“ANI-2x Data Set Guidelines”).

Notes

Upstream energies are in Hartree, forces in Hartree/angstrom, and positions in angstrom. By default energies/forces are converted to eV and eV/angstrom (the convention used elsewhere in xnn); units="hartree" keeps the raw values.
Only the wB97X/6-31G(d) release is used: it is the level ANI-2x was fit to and the only released ANI-2x level that ships usable forces (the def2-TZVPP files omit forces upstream).
The one ~3.7 GB archive is downloaded once (cached and MD5-verified) and extracted. Use n_atoms to select atom-count groups and max_conformations to cap the amount materialised – the full set is ~9.6 M conformations and will not fit in memory at once.
ANI-2x ships no official split. split in {"train", "val", "test"} applies a per-conformation 80/10/10 partition with a fixed seed, so splits are disjoint and reproducible; split=None returns {"all": ...}.

Classes

ANI2xBuilder()

Builder for the ANI-2x data set (7-element wB97X energies & forces).