Aurora
The curated chemistry database for virtual screening.
833,593 compounds under a single append-only identity, curated from 936,635 raw records across eight licence-cleared sources. Gated 3D geometry, ADMET, fingerprints and docking-ready files — every structure carrying its origin.
Get first access when v1 ships
Core capabilities
Virtual screening
Similarity and substructure search over ECFP, MACCS, RDKit and e3fp fingerprints, computed for every gate-approved compound.
Docking-ready
787,666 compounds shipped with SDF, PDBQT and checksum. Benchmarked against PDB 7A47 on AutoDock-GPU and Vina.
ML datasets
RDKit 1D/2D descriptors at 100% coverage; 3D shape, e3fp and ADMET-AI profiles across the 94.6% that clear the geometry gate.
Natural products
COCONUT, BrNPDB, BraCoLi and TCMBank curated side by side — including Brazilian and traditional Chinese medicine collections.
Verified against production
v1.7.0 · 2026-08-14The funnel
Nothing advances a stage without passing its gate.
Every figure below was read live from the production databases, not copied from an exported file. Where a residual discrepancy exists, it is reported rather than reconciled silently.
- 01—833,593
Curated universe
Append-only record. One identity per compound, never reissued.
- 0298.2%818,700
Reached conformation
Canonical row in conformers — 3D geometry generated.
- 0394.56%788,214
Chemistry gate v1.1.0
Geometry validated. 30,486 failed the gate and are flagged, not dropped.
- 0494.50%787,666
Distributable
SDF, PDBQT and checksum synced on disk and in the database.
The databases
Eight sources. Every licence cleared before a single record was read.
Licensing is a gate, not a footnote: a source with no cleared status gets no identity. BindingDB and full ChEMBL 37 are measured and deferred — read only, never ingested.
| Source | Raw records | Format | Licence |
|---|---|---|---|
| COCONUT | 738,827 | CSV + SDF | Cleared · open natural products |
| NCI | 92,370 | SDF | Cleared · US federal public domain |
| TCMBank | 61,965 | XLSX + MOL2 | Cleared · traditional Chinese medicine |
| Drug Repurposing Hub | 22,612 | TSV | Cleared · non-commercial, Broad Institute |
| BrNPDB | 15,651 | MOL2, native 3D | Cleared · direct consent, UNESP/NuBBE |
| AntibioticDB | 3,513 | CSV | Cleared |
| BraCoLi v1 | 1,176 | SDF | Cleared |
| Iracema | 521 | CSV | Cleared |
Deferred, not ingested — BindingDB (3,234,499 rows measured) and ChEMBL 37 (2,897,819 records) are read-only until a licensing decision is made.
Method
The gate is the product.
Standardisation
ChEMBL Structure Pipeline: Checker, Standardizer, GetParent. Salt strip, tautomer canonicalisation, charge normalisation, InChIKey reconciliation.
Geometry
3D conformers generated and validated against chemistry gate v1.1.0. PoseBusters: 94.31% raw, 98.80% reweighted for the real macrocycle rate.
Provenance
833,621 compound–source links across eight sources. A molecule deposited by two sources keeps both links; identity is never merged away.
Early access to Aurora.
Aurora is internal today. Join the list and you hear from us when the first versioned release ships — file download and a versioned API, in that order.