Downloads

Three ways to take the data: single files on every protein page (Downloads section); your selection as ZIP, CSV or a download script (tick pairs in any table); and the bulk release below.

Bulk release

Release v1.1. Exact sizes; checksums in MANIFEST.json.

File	Content	Size
`tables/compound_entries.parquet`	Every catalogue number of the library: docked structure (PVL), what was docked, desalted entries	430 KB
`tables/compound_entries_annotated.parquet`	Catalogue numbers with names, ChEMBL IDs, ATC classes and synonyms	354 KB
`tables/compounds.parquet`	Ligand IDs (PVL): one per docked structure, with its SMILES and InChIKey	196 KB
`tables/modes.parquet`	Every Vina mode: affinity, RMSD bounds, byte position of its pose	35.8 GB
`tables/open_targets.parquet`	Open Targets diseases of each protein	1.7 MB
`tables/pair_pvl.parquet`	Protein-ligand pairs: the canonical run of each pair	54.6 MB
`tables/pairs_by_compound.parquet`	Pair results sorted by compound: highest affinity, pocket, mode, PoseBusters, ChEMBL evidence	3.3 GB
`tables/pairs_by_protein.parquet`	The same pair results sorted by protein	3.2 GB
`tables/pocket_runs.parquet`	One row per pair and pocket: best affinity, number of modes, byte position of the mode block	13.5 GB
`tables/pockets.parquet`	P2Rank pockets of each protein: score, probability, centre, residues	3.0 MB
`tables/posebusters.parquet`	PoseBusters checks of each highest-affinity pose	55.1 MB
`tables/proteins.parquet`	Proteins: identifiers, names, classes, model quality	3.7 MB
`tables/pvl_flags.parquet`	ChEMBL evidence flags per protein and ligand ID	46 KB
`tables/replicate_runs.parquet`	Pairs docked again under another catalogue number of the same structure (kept, not shown)	48.3 MB
`poses/`	One PVATLAS1 pose file per protein: highest-affinity poses and all modes of each pocket run	1.83 TB
`poses_redock/`	One PVATLAS1 pose file per protein for the desalted forms of salt entries	601.2 GB

Step by step, with a Python reader of the pose files and commands for PyMOL, ChimeraX and Jupyter: Download and visualise.

One pose without downloading the whole file

Every protein has one pose file, poses/<UniProt>.poses (and poses_redock/<UniProt>.poses for the desalted forms of salt entries). The protein's table, proteins/<UniProt>.json.gz, gives the byte offset and length of each row's highest-affinity pose (columns top_off, top_len; column pf names the file). One HTTP range request returns one gzip member:

# top_off and top_len of a row of https://bimsbstatic.mdc-berlin.de/akalin/asarigun/pocketvina_atlas/data/v1.1/proteins/P00519.json.gz
curl -r <top_off>-<top_off+top_len-1> https://bimsbstatic.mdc-berlin.de/akalin/asarigun/pocketvina_atlas/data/v1.1/poses/P00519.poses | gunzip > pose.sdf

The mode blocks (all poses of one pocket run) are found the same way through the file's table B; the format is described in MANIFEST.json and the table modes.parquet lists every mode with its byte position.

Query the tables without downloading them

duckdb -c "SELECT uniprot, catalog_id, vina, best_pocket, pb_status
           FROM 'https://bimsbstatic.mdc-berlin.de/akalin/asarigun/pocketvina_atlas/data/v1.1/tables/pairs_by_compound.parquet'
           WHERE catalog_id = 'S1026' AND pb_status = 'valid'
           ORDER BY vina LIMIT 20"

Verify

MANIFEST.json in the release folder lists every file with its size and SHA-256.

Licence

The licence of the atlas data has not been decided yet. It will be stated here before the first public release.

The AlphaFold DB models are CC BY 4.0. The redistribution terms of the compound library are being checked.