AntiSite: Unified paratope prediction for antibodies and nanobodies with or without structure

Antisite benchmark structures and split metadata Source code at: https://github.com/aggelos-michael-papadopoulos/AntiSite Paper (currently on Biorxiv): https://doi.org/10.64898/2026.08.24.745914 This dataset contains processed antibody–antigen coordinate inputs and split metadata for AntiSite, covering Paragraph-expanded, PECAN, MIPE and AACDB, together with the final SAbDab2 paired-antibody and VHH nanobody test sets. The complete v2 archive contains 7,625 dataset entries and 15,250 processed PDB files, provided in a single ZIP. Version v2 adds 178 SAbDab2 test complexes and 140 genuine VHH–antigen complexes representing 104 distinct antibodies. The original four-benchmark data remain unchanged from v1. Published split assignments are retained for Paragraph-expanded, PECAN and MIPE, while AACDB uses the cluster-disjoint split documented by AntiSite. The archive includes antibody and antigen sequences, chain assignments, source identifiers, provenance and licensing information, a unified per-complex manifest, and SHA-256 checksums. Antigen coordinates support reference-contact construction and are not AntiSite model inputs.