About PocketVina Atlas
Each pocket a lock, each drug a key.
A resource of the Bioinformatics and Omics Data Science Platform (Akalin Lab), Berlin Institute for Medical Systems Biology (BIMSB), Max Delbrück Center, Berlin.
The people behind it: Team.
PocketVina
Every pose in the atlas was computed with PocketVina (package pocketvina-gpu; PocketVina 0.1.0; desalted forms of salt entries: PocketVina 0.2.0), a docking method that conditions on several pockets at once: P2Rank predicts the pockets of a protein structure, and QuickVina2-GPU docks the compound in each of the top pockets, so a drug is not forced into one guessed site. Paper: PocketVina Enables Scalable and Highly Accurate Physically Valid Docking through Multi-Pocket Conditioning (Sarigun A, Uyar B, Franke V, Akalin A, arXiv:2506.20043).
The name of the method is also the name of the atlas: each pocket a lock, each drug a key. It follows the lock-and-key model that Emil Fischer proposed in Berlin in .
The screen
The compounds of an approved-drug library (Selleck, 3,108 catalogue entries) were docked against 18,713 human proteins (AlphaFold DB v6 models): QuickVina2-GPU in 25 Å boxes centred on the top 5 P2Rank pockets of each protein, 20 modes per pocket run. That gives 71,289,971 protein × compound pairs with a result and 6.6 B poses; the docked structures hold 2,559 distinct drug molecules.
The library holds approved drugs and other active ingredients listed in pharmacopoeias. Some entries are veterinary drugs, natural products or reagents rather than human drugs; the approval status of each compound comes from ChEMBL (max phase).
Highest-affinity pose and PoseBusters
For each protein-drug pair, the highest-affinity pose (the lowest Vina score) over all pockets and modes was written as SDF and checked with PoseBusters (21 checks of chemistry, geometry and clashes with the protein). Other modes were not checked, and the viewer says so. A timeout or a PoseBusters error means the check did not finish; it is not a failure. PoseBusters was not run on the desalted forms of salt entries (Salts): their rows say "not run".
Reading the scores
Vina scores are docking scores, not measured affinities. Large ligands score better almost everywhere, so compare a drug with other drugs on the same protein, and a protein with other proteins for the same drug. Each page shows both ranks.
Salts
Some library entries are salts, written with the counter-ion first. As screened, such an entry was docked as that first fragment (a salt counterion, a solvent or one component of a product), not as the drug: 31 ligand IDs hold such structures. Their rows keep their scores and poses, labelled with what was docked, and atlas-wide statistics count drug entries only. For 793 of these salts the atlas also has the desalted form: the largest fragment, docked as the drug with PocketVina 0.2.0, under the key <catalogue number>-desalted. The drug's page lists both forms and links them. PoseBusters was not run on the desalted forms; their rows say so.
ChEMBL evidence
MoA: the pair is a ChEMBL mechanism of action, a known target of the drug. Assay: strict binding-assay evidence. Weak: permissive assay evidence only. No badge means no ChEMBL evidence, which is not the same as no binding: most pairs were never tested.
Associated diseases
The diseases on a protein page come from the Open Targets Platform 26.09 (data under CC0 1.0): the ten diseases with the highest overall association score for the protein's gene, as Open Targets lists them on the gene's page (direct associations, measurements left out). The score combines genetic, drug, literature and other evidence; it is not related to the docking results. Paper: Buniello A, et al. Open Targets Platform: facilitating therapeutic hypotheses building in drug discovery. Nucleic Acids Research 53, D1467 (2025).
Numbers
Every number on this site is read from meta.json of the data release, together with its definition and the table it was computed from. Release v1.1.
| Number | Value | Definition | Source |
|---|---|---|---|
| Proteins | 18,713 | Human proteins with docking results: AlphaFold DB v6 models, one per UniProt accession. | tables/proteins.parquet |
| Library entries | 3,108 | Catalogue entries of the approved-drug library (Selleck) used as input. | the library list of the screen (Selleck catalogue) |
| Catalogue entries docked | 3,859 | Catalogue entries docked: library entries whose ligand was prepared and docked, and the desalted forms of salt entries (<CAT>-desalted). | tables/compound_entries.parquet |
| Catalogue entries with results | 3,830 | Catalogue entries with docking results, desalted forms included. | tables/compound_entries.parquet |
| Desalted forms | 793 | Desalted forms of salt entries: the salt's largest fragment, docked as the drug; key <CAT>-desalted. | tables/compound_entries.parquet |
| Desalted forms with results | 793 | Desalted entries with docking results. | tables/compound_entries.parquet |
| Salt entries without a desalted form | 25 | Salt entries without a desalted form: the largest fragment is again a counterion, a solvent or a product component, or the ligand could not be prepared. | excluded.tsv |
| Drug molecules | 2,559 | PVL IDs (docked structures) with results whose docked structure is the drug molecule. | tables/compounds.parquet (docked_as = drug) |
| Non-drug entries | 31 | PVL IDs with results whose docked structure is a salt counterion, a solvent or a product component, not the drug. | tables/compounds.parquet (docked_as != drug) |
| Not docked as the drug | 22 | Catalogue numbers whose drug molecule was not docked in this release (a counterion, solvent or product component was docked instead, and the drug has no other docked entry, desalted entries included). | not_docked_as_drug.csv |
| Pairs (all) | 71,289,971 | All protein × compound pairs with a docking result: every catalogue entry, both forms of salt entries and non-drug entries included. | pairs + replicate_runs (meta.json) |
| pairs_screen | 56,450,562 | Protein × compound pairs of the library entries as screened: one per protein and catalogue number. | meta.json: all pairs minus the pairs of the desalted forms |
| pairs_desalted | 14,839,409 | Protein × compound pairs of the desalted forms of salt entries. | meta.json: pairs of the desalted forms |
| Unique molecule pairs | 48,406,051 | Protein-PVL pairs: one canonical run each (the lowest as-screened catalogue number docked as the drug, else the lowest desalted form, else the lowest catalogue number). PoseBusters checked the highest-affinity pose of the as-screened runs; it was not run on the desalted forms. | tables/pair_pvl.parquet (canonical) |
| Drug pairs | 47,826,043 | Pairs of drug PVLs (docked_as = drug). | tables/pair_pvl.parquet |
| Replicate runs | 22,883,920 | Pairs docked again under another catalogue number of the same docked structure; kept, not shown. | tables/replicate_runs.parquet |
| Pocket runs | 341,234,334 | Docking runs, one per pair and pocket (up to 5 pockets per protein), over all catalogue numbers. | tables/pocket_runs.parquet |
| Pocket runs without all modes | 571,651 | Pocket runs whose all-mode poses are not in this release (scores and highest-affinity pose are). | missing_modes.json |
| Proteins with missing modes | 105 | Proteins with at least one pocket run without all-mode poses. | missing_modes.json |
| poses_screen | 5,193,646,087 | Docked poses (Vina modes) of the library entries as screened. | meta.json: all poses minus the poses of the desalted forms |
| poses_desalted | 1,401,250,938 | Docked poses (Vina modes) of the desalted forms of salt entries. | meta.json: poses of the desalted forms |
| Poses | 6,594,897,025 | Docked poses (Vina modes) with scores over all pocket runs. | tables/modes.parquet |
| Pockets per protein | 5 | P2Rank pockets docked per protein (the top-ranked ones; fewer if P2Rank finds fewer). | PocketVina settings |
| Modes per pocket run | 20 | Vina modes requested per pocket run (Vina may return fewer). | PocketVina settings |
| Docking box (Å) | 25 | Edge of the cubic docking box around each pocket centre, in Å. | PocketVina settings (Vina configs) |
| PoseBusters checks | 21 | PoseBusters checks run on the highest-affinity pose of each pair (not run on the desalted forms: pb is null). | bust_results.csv |
| Batches in the screen | 1,845 | Batch folders in the screen. | results/ |
| Batches with docking | 1,758 | Batch folders with docking output (the others had no protein with pockets). | phase-2 import reports |
Security and privacy
The site is static files. It has no accounts, cookies, uploads or tracking, and it loads nothing from other sites. The 3D viewer (Mol*) runs in an isolated frame with its menus and file loading turned off. Your selection stays in this browser tab.