API

Every page is built from static files. Fetch them directly; there is no key and no rate limit. Paths are relative to https://bimsbstatic.mdc-berlin.de/akalin/asarigun/pocketvina_atlas/data/v1.1/.

"open_targets": {
  "release": "26.09",               Open Targets Platform release
  "ensembl_id": "ENSG00000146648",  the Open Targets target (Ensembl gene) of the UniProt accession
  "association": "direct",          direct or indirect associations
  "diseases": [                     top 10, highest score first
    {"id": "MONDO_0005233", "name": "non-small cell lung carcinoma", "score": 0.8465},
    {"id": "MONDO_0005061", "name": "lung adenocarcinoma", "score": 0.7754},
    …
  ]
}
# links: https://platform.opentargets.org/evidence/<ensembl_id>/<id>, https://platform.opentargets.org/target/<ensembl_id>/associations
# score: overall association score, 0–1 (not related to docking results)

Examples

# the ten ligands with the highest affinity on EGFR (jq)
curl -s https://bimsbstatic.mdc-berlin.de/akalin/asarigun/pocketvina_atlas/data/v1.1/proteins/P00533.json.gz | gunzip | jq -r '.rows | sort_by(.[1]) | .[:10][] | @tsv'

# one highest-affinity pose with a range request (top_off, top_len of its row; pf = 1: poses_redock/)
curl -s https://bimsbstatic.mdc-berlin.de/akalin/asarigun/pocketvina_atlas/data/v1.1/proteins/P00519.json.gz | gunzip | jq -c '(.columns | index("pvl")) as $i | .rows[] | select(.[$i] == "PVL00129")'
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

Identifiers and file names

Things that exist outside the atlas keep their own identifiers (UniProt, AlphaFold); the one identifier the atlas adds is the ligand ID, because about a third of the compounds have no ChEMBL ID. Every identifier is shown on its page with a Copy button; open any of them with /id/?<ID> (for example /id/?P00519_PVL00129_pk1_m01).

Saved files are named after these identifiers, as <ID>_<content>.<type>; the AlphaFold model keeps AlphaFold's name, and the tool that made a file (P2Rank, QuickVina2-GPU, PoseBusters) is in its metadata, not in its name:

AF-P00533-F1-model_v6.pdb           the AlphaFold model, unchanged
P00533_pockets.json                 the docked pockets: centre, residues, P2Rank scores
P00533_poses.poses                  every pose of the protein (PVATLAS1: tables of byte offsets, then one gzip member per pose)
P00533_poses_redock.poses           the poses of the desalted forms on the protein (same format)
P00533_compounds.csv                one row per compound on the protein
P00533-pk4_residues.csv             the residues of one pocket
P00519_PVL00129_top.sdf             one pair: its highest-affinity pose (the one PoseBusters checked)
P00519_PVL00129_pk1_modes.pdbqt     one pair, one pocket: all its modes (MODEL k is pose P00519_PVL00129_pk1_m<k>)
P00519_PVL00129_modes.csv           one pair: the Vina scores of all its modes in all pockets
P00519_PVL00129_posebusters.json    the PoseBusters checks of that pose
PVL00129_proteins.json.gz           one drug: its docked proteins
PVL00129.svg                        one drug: its 2D structure

No file name carries a version. The release is in the folder of a selection (PocketVina-Atlas_<release>_selection_<date>_<n>pairs/, with the release label such as v1, or the data version of a preview: v1.1 in this build) and in its MANIFEST.json, which also gives each pose file's source and byte range. Pose files are the release's records byte for byte, so they carry no added fields. On the server the files keep their paths (the table above); only the names files are saved under follow the scheme.

best in field names (best_pocket, best_mode, best_affinity) means highest-affinity: the pose with the lowest Vina score. It is not a judgement of quality or correctness; the names stay as they are so that scripts keep working. Saved files say top, as do top_off and top_len.

Earlier identifiers

Earlier previews used longer identifiers and file names, with a PV prefix and a version. They still work: the search box and /id/?<ID> take them to the same page.

Compressed files

The JSON files under search/, proteins/ and compounds/ are published gzip-compressed only (.json.gz); the server sends them as they are, so curl -s URL | gunzip gives the JSON. The pages unpack them in the browser.

Canonical page URLs: /protein/<UniProt>/, /drug/<PVL ID>/ (/drug/<catalogue number>/, /drug/<catalogue number>-desalted/, /drug/<ChEMBL ID>/ and /drug/<name>/ redirect there), /proteins/<class>/, /drugs/<ATC code>/; /id/?<ID> opens the page of any atlas identifier.