AMES_xTB: GFN2-xTB quantum-chemical descriptors for the Ames mutagenicity dataset

AMES_xTB provides semi-empirical quantum-chemical descriptors (GFN2-xTB) for 7,251 compounds from the Ames mutagenicity benchmark of the Therapeutics Data Commons (TDC). The data are split into two linked tables: a molecule-level table with the SMILES, the Ames label and global descriptors (frontier-orbital energies, electrophilicity index, total energy, total polarizability), and a token-level table with the SMILES split into MolEncoder tokens, where each token carries the local (atom-level) descriptor values mapped onto it. These are the descriptor values used to train the descriptor-augmented MolEncoder models in the accompanying study. All values are reported as parsed from the GFN2-xTB output files, with no standardisation or rescaling. Column definitions and units are given in README.md.