COLUMN_GUIDE.md

Column guide

data/models.parquet

Identity and files

Columns	Meaning
model_id	Stable model key.
method_id, method, source_group	Method family, display name, and source tranche.
ligand_id, seed_ordinal, native_seed, sample_index, native_sample_index	Model grid identifiers.
structure_archive, structure_member	Archive and member containing the aligned CIF.
coordinate_sha256, aligned_structure_sha256	Native-source and aligned-CIF hashes.
source_coordinate_format, source_coordinate_bytes, aligned_structure_bytes	Coordinate format and sizes.

Confidence and geometry

Columns	Meaning
protein_ligand_iptm, iptm_source_key	Protein–ligand iPTM and the engine-native field used.
dist_<RES1>_<RES2>_sidechain_centroid_a	Side-chain centroid distance, Å.
dist_<RES1>_<RES2>_ca_a	Cα distance, Å.
alignment_rmsd_to_1nrl_a	RMSD of the governed 162-Cα alignment.
alignment_rotation_json, alignment_translation_json	Rigid transform applied to the source model.
model_ligand_centroid_[xyz]	Ligand centroid in the aligned frame.
ligand_centroid_distance_to_truth_a	Truth-relative ligand-centroid distance, Å.
protein_strict_core_rmsd_to_truth_a	Truth-relative receptor-core RMSD, Å.

Official OpenStructure fields

official_ost_lddt_pli, official_ost_lddt_lp, official_ost_bisyrmsd_a, and official_ost_bb_rmsd_a are official scorer outputs. The adjacent status, state, state_label, contact-count, mapped-residue, chain-mapping, route, and graph-normalization columns describe scorer coverage and provenance.

Prepared chemistry and contacts

Columns	Meaning
ligand_state_status	Generation-state provenance status.
chemistry_feature_status, chemistry_feature_error	Prepared-top chemistry feature outcome.
ligand_mapping_status, ligand_mapping_mode	Authoritative graph-to-coordinate mapping outcome.
authoritative_*_bond_distance_a, authoritative_bad_bonded_distances	Bonded-distance mapping audit.
ligand_pose_mmff_*, ligand_strain_*, strain_baseline_energy_kcal_mol	MMFF94s pose energy and conformer-baseline strain.
prolif_status, prolif_protocol, prolif_total_interactions, prolif_contact_residue_count	ProLIF outcome and totals.
prolif_count_<interaction>	Count by interaction type.
prolif_contacts_json	Residue-level interaction counts.

PoseBusters summary

posebusters_result_status, posebusters_pass_all_configured_checks, posebusters_n_failed, and posebusters_failed_checks summarize the run. The remaining PoseBusters boolean columns are individual checks: true is pass, false is fail, and null means not run.

data/posebusters_full.parquet

This contains the raw full-report diagnostics plus staging provenance.

Column group	Meaning
mol_pred_sha256, mol_cond_sha256	Exact staged ligand and receptor hashes.
authoritative_prepared_smiles, authoritative_formal_charge, atom_assignment_mode	Topology and atom mapping used for staging.
authoritative_min_bond_distance_a, authoritative_max_bond_distance_a	Mapped ligand bond-length extremes, Å.
ligand_coordinate_max_abs_delta_a, receptor_pdb_rounding_max_abs_delta_a	Coordinate changes introduced by staging.
mol_pred_loaded through volume_overlap_with_waters	PoseBusters pass/fail checks.
number_*, shortest_*, longest_*, most_extreme_*	Raw bond, angle, clash, planarity, and contact diagnostics.
mol_pred_energy, ensemble_avg_energy, energy_ratio	Internal-energy diagnostics.
smallest_distance_<partner>, volume_overlap_<partner>	Raw protein/cofactor/water proximity diagnostics.
posebusters_*	Derived status, failed-check list, software version, and protocol.

Lookup tables

data/ligands.parquet contains ligand SMILES, prepared state, charge, molecular weight, and class. data/truth_structures.parquet contains reference hashes, archive members, alignment fields, ligand-centroid fields, and the strict-core coordinates used for truth-relative comparisons.