isalgo/vdjdb_structure_models

Predicted structures for VDJdb records

This repository contains structure data for selected VDJdb records obtained using AI-based modelling in data/ folder:

• pdb_files.tgz contains predicted TCR:pMHC structures with canonical chain names, orientation and placement, superimposed by aligning and rotation. File names start with tcr_pmhc_hash which must be used for connecting the structure with the VDJdb record.
• pdb_files_native.tgz contains experimental TCR:pMHC structures from PDB, processed to canonical coordinates in the same way as pdb_files.tgz.
• coordinates_aa.tgz contains tables with amino acid Ca atom coordinates for each residue of deposited structures.
• contacts_aa.tgz contains files with residue pairs at a distance <= 5A for each of the deposited structures. Currently only TCR alpha CDR3-peptide and beta CDR3-peptide are considered.
• complementarity_maps.tgz and complementarity_maps_simplified.tgz contains 2D projections of TCR alpha CDR3, TCR beta CDR3 and peptide residue coordinates in for of SVG plots produced by matplotlib.

Metadata for all structures is concatenated with corresponding VDJdb record metadata and stored in vdjdb_structures_metadata.tsv.gz. The file contains default VDJdb "full table" columns together with the following columns calculated for each structure:

• num_contacts - number of contacts in the available structure between TCR and peptide.
• ranking_confidence, plddt, ptm, iptm, tcr_pmhc_iptm - TCRmodel2 quality metrics (for more information please visit TCRmodel2 webpage).
• scanning_angle and pitch_angle - scanning and pitch TCR:pMHC structure angles calculated with STCRpy.
• is_native - whether the structure is experimentally acquired (from PDB) or predicted.
• tcr_pmhc_hash - unique TCR:pMHC record hash for matching predicted structure PDBs.

Contents at a glance

metadata rows	15,018 (11,234 distinct tcr_pmhc_hash)
predicted / native rows	14,612 / 406
PDB files shipped	10,884 predicted (pdb_files.tgz) + 369 native (pdb_files_native.tgz)
epitopes / MHC alleles	249 / 81
MHC class	14,915 MHC-I, 103 MHC-II
species	14,928 human, 90 mouse

A metadata row without a matching PDB is normal - the file set is the subset that modelled and passed QC. Join on tcr_pmhc_hash; predicted filenames are aligned_aligned_<hash>.pdb, natives are aligned_<pdbid>.pdb.

Provenance

• TCR:epitope pairings are EXPERIMENTAL - VDJdb records, carried through with the full VDJdb "full table" columns (reference.id, method.*, meta.*, vdjdb.score).
• Predicted structures are COMPUTED (TCRmodel2/AlphaFold). ranking_confidence, plddt, ptm, iptm, tcr_pmhc_iptm are the generator's own confidence metrics, not measurements.
• Native structures are EXPERIMENTAL (RCSB), re-processed into the same canonical frame.
• Derived columns (num_contacts, scanning_angle, pitch_angle) are computed from the deposited coordinates — for predicted rows they inherit the model's error, and the angles in particular track AlphaFold confidence rather than being independent of it.

Used by

The TCRen2 receptor-ranking benchmark (isalgo/tcren_structures) draws its real-binder positives from this pool - filtered to is_native == False, restricted to hashes present in pdb_files.tgz, and with TCRvdb pairs excluded - then matches them per epitope against the decoys in isalgo/tcren_structures::vdjdb_negatives/ and ::immrep23_negatives/. See isalgo/tcren_structures::vdjdb_binder_benchmark/README.md.

Links to current VDJdb records

Current record metadata and complete structure accounting link 8429 of the 11253 deposited PDB structures to current record IDs. The metadata marks exact matches to submitted junctions, repaired junctions, or both, and keeps observation/accession links with sequence differences explicit. All remaining structures have recorded dispositions; original structure files and source metadata are unchanged. For a current positive collection, select through the current record links and the required match flags; presence in a historical archive alone does not establish a retained positive assignment. The original mixed archives still include the withdrawn assignments preserved in low_confidence/withdrawn_10x/.

Of the 2811 structures in the low-confidence withdrawn-10X collection, 2538 (90.3%) are assigned to KLGGALQAK. The other 273 comprise GILGFVFTL (117), RAKFKQLL (95), and six other peptides (61 combined).

Native source review and PDB coverage

The native review accounts for all 14 previously unlinked native complexes: 9GV6 now has current record ID VDJDB0000339155; the other 13 were already excluded chemical-ligand observations and remain preserved with their literal chemistry. Coverage metadata confirms coordinates for all 360 declared PDB_Database accessions. The 18 missing accessions were downloaded; complete mmCIF sources, including the 14 reviewed complexes, are in data/pdb_sources. Current metadata retains original/repaired match flags and explicit mismatch/exclusion reasons.

Database builds, identifiers and downloads

The curation source and build are antigenomics/vdjdb-db. Build-output documentation describes all three database bundles, their units and filters, compact Parquet/gzip tables, and the structure joins. Coordinates remain in this dataset; database bundles contain evidence IDs and supplied metrics. A database build does not copy these archives into every format.

Use metadata/current/record_links.tsv.gz to join to a database build's records.parquet on record_id; join chains.parquet on the same ID for TRA/TRB sequences. The exact database input revision and SHA256 values are in metadata/current/verification.json. Read the current metadata contract before filtering original/repaired matches or resolving many-to-many archive links. idx, complex.id and the historical source hash are not substitutes for the current record_id.

Location	Purpose and format
vdjdb_structures_metadata.tsv.gz	Preserved summer-release source metadata
data/*.tgz	Preserved canonical PDBs, residue coordinates, contacts and SVG maps
metadata/current/	Gzip TSV links, source/asset dispositions and verification manifest
metadata/native_review/	Fourteen reviewed native fates and coverage of declared PDB accessions
data/pdb_sources/	Supplemental complete deposited mmCIF sources, gzip-compressed, with hashes
low_confidence/withdrawn_10x/	Withdrawn-assignment metadata and byte-verified selected assets
tools/	Regeneration and verification commands; see each script's --help

The low-confidence archives deliberately retain selected files also present in the historical mixed archives, providing a usable standalone partition while preserving the original collection. Download only the archive types needed for an analysis. Current links and dispositions are flat gzip TSV rather than uncompressed or repeatedly joined text copies. The expanded current metadata is a convenience view; record_links.tsv.gz is the smaller association table. No new structure prediction or resampling was needed for this migration.

For strict paired-junction/peptide matching, use matches_original or matches_repaired. For agreement on that pair plus recorded species and both MHC chains/class, require exact_tcr_pmhc_match. link_basis distinguishes historical repair and accession provenance from literal sequence matches. Association does not establish identical full receptor constructs. The cdr3 fields use VDJdb's anchor-inclusive junction convention, not anchor-excluding CDR3. Withdrawn assignments have experimental_negative=false; their receptors can have valid observations against other peptides. The independent audit found no exact paired-junction/peptide match to any current positive record for these 2,811 structures, across both original and repaired states. Receptors of 174 hashes occur against other peptides, as recorded in the audit files.

Current-record coordinate proofreading

The independent VDJdb proofreading script now checks PDB accessions across every current record, including observations retained under publication chunks after deduplication. It extends the earlier 360-accession aggregate-chunk check to 367 accessions across 411 current records. Six additional deposited files (9NU7, 9NV6, 9NV2, 9NVE, 9NW2, 9NVA) are preserved under data/pdb_sources/; 7NA5 was already in the native archive. The script verifies all archived assets, source checksums and current record links, and writes separate coordinate-to-record and native-coverage Parquet tables. The existing historical archives and source-row associations are unchanged. Run processing/proofread_structure_files.py in the VDJdb source repository; see the structural output contract.

Supplemental coordinate catalogue

The supplemental catalogue adds content-addressed coordinates and reviewed current-record links. It separates reported assignments, constructed mismatches, mutants and native structures. A coordinate file or sequence match alone does not establish experimental validation. Generator confidence values remain separate from assay outcomes; no assay outcomes are included.