RASCAL MCES pairwise similarity for MassSpecGym compounds (pilot, 43.5%)

Exact pairwise Maximum Common Edge Subgraph (MCES) similarity for 31,587 compounds from MassSpecGym, computed using the RASCAL algorithm (RDKit implementation) with similarity threshold 0 and per-pair timeout of 600s. This is a partial release covering 216,900,000 of 498,792,891 total pairs (43.5%). Timed-out pairs (0.01%) are flagged, not discarded. The remaining pairs are not witholded, they are just not computed yet. Files: - pairs.parquet (2.1 GB): zstd-compressed Parquet with columns cid_a, cid_b, similarity, timed_out, compute_time_ms - compounds.parquet (1.1 MB): CID, InChIKey, SMILES, HeavyAtomCount for all 31,587 compounds - metadata.json: RDKit version, timeout, total pairs, chunk count The similarity distribution has mean 0.241, median 0.227, and standard deviation 0.136. For details see: https://github.com/LucaCappelletti94/pubchem-rascal-mces