# ChEMBL 37: curated bioactivity database of drug-like molecules

## Facts

- Provider: EMBL-EBI (ChEMBL group)
- License: CC BY-SA 3.0
- Format: FASTA, HDF5 (in sampled files)
- Stored at: Amazon S3, s3://scigantic-chembl (us-east-1)
- Access: public, readable without an account
- DOI: https://doi.org/10.6019/CHEMBL.database.37
- Category: Drug Discovery
- Source: https://www.ebi.ac.uk/chembl/
- Page: https://scigantic.com/archive/97a98fc5-e6f9-555c-a22f-f2b548775f14
- Keywords: Drug Discovery, FASTA, HDF5, AWS Open Data

## About

ChEMBL is EMBL-EBI's manually curated database of bioactive molecules with drug-like properties, and the reference source for structure-activity data extracted from the medicinal chemistry literature. This is release chembl_37, prepared 2026-05-01: 2,921,148 distinct compounds (2,897,819 with mol files), 3,824,604 compound records, 24,527,044 activity measurements, 1,970,438 assays, 18,552 targets and 101,100 source documents. Beyond the primary literature the release aggregates roughly 50 deposited sets, among them PubChem BioAssays, BindingDB patent bioactivity, SureChEMBL patent data, DrugMatrix, the Gates Foundation compound collection, the EUbOPEN chemogenomic library, GSK's published kinase inhibitor sets and a long tail of malaria, tuberculosis and schistosomiasis screening campaigns. This archive is a mirror of the upstream release, mounted read-only at ~/archive so files stream on demand rather than being copied at launch. Everything sits under chembl_37/. sqlite/chembl_37.db is the complete relational database exactly as EBI ships it, already expanded from the 5.4 GB loader tarball so there is nothing to download or unpack. parquet/ holds the same 72 tables re-encoded one file per table, 1.66 GB against the database's 30.5 GB, with _index.json and _index.txt listing every table's row count, byte size and column types. bulk/ holds the artifacts people use without SQL: the SDF structure file, the chemreps table of SMILES, InChI and InChI keys, 2048-bit radius-2 Morgan fingerprints in both FPS text form and the FPSim2 .h5 database, target and biotherapeutic FASTA files, the HELM monomer library and the ChEMBL-to-UniProt target mapping. The upstream README, LICENSE, REQUIRED.ATTRIBUTION, checksums.txt, release notes and schema documentation sit at the top of the release directory. Use the parquet copy for joins. Measured cold on the real mount in a Standard-tier notebook, the five-table potency query in the schema card takes 7.3 seconds over parquet and does not finish in 300 seconds against SQLite, and aggregating over the molfile column is 5.9 seconds against 50.5. Single-table work is a different story: SQLite ships 20 indexes on the activities table alone, so an indexed count is 2.0 seconds there against 5.5 in DuckDB, and even a full single-table scan finishes in about 11 seconds. Both engines return identical answers, so the choice is about shape of query, not correctness. ChEMBL also publishes a REST API at https://www.ebi.ac.uk/chembl/api/data/ with an official Python client (chembl-webresource-client), which remains the better choice for fetching a handful of records against the newest release rather than analysing the corpus. Useful for target-based activity mining, QSAR and ADMET dataset construction, chemical similarity and substructure search, drug repurposing questions, and as the bioactivity backbone that most public medicinal chemistry benchmarks are cut from. ChEMBL data is licensed CC BY-SA 3.0, which permits this redistribution; attribution requires naming the release, which is why the release number is the top-level directory. See LICENSE, REQUIRED.ATTRIBUTION and _SCIGANTIC_MIRROR.txt in the archive.

## Load the data

```python
import os
import gzip
opener = gzip.open if os.path.join(os.environ['SCIGANTIC_MOUNT_PATH'], 'chembl_35/bulk/chembl_35.fa.gz').endswith('.gz') else open
with opener(os.path.join(os.environ['SCIGANTIC_MOUNT_PATH'], 'chembl_35/bulk/chembl_35.fa.gz'), 'rt') as fh:
    for _ in range(8):
        print(fh.readline().rstrip())
```

Run in a Scigantic notebook on 2026-11-19 against `chembl_35/bulk/chembl_35.fa.gz`: FASTA: first record header '> CHEMBL4523954 [A0A024AXB9] Nonstructural protein 3 (Zika virus)', read length 60.

From any machine:

```python
import boto3
from botocore import UNSIGNED
from botocore.client import Config
s3 = boto3.client('s3', region_name='us-east-1', config=Config(signature_version=UNSIGNED))
s3.list_objects_v2(Bucket='scigantic-chembl', Prefix='', Delimiter='/')
```

Lists the files with anonymous access. Runs on any machine, with no cloud account.

## Data structure

File types in sampled files: FASTA, HDF5

Example file: `chembl_35/bulk/chembl_35.fa.gz`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/97a98fc5-e6f9-555c-a22f-f2b548775f14
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/97a98fc5-e6f9-555c-a22f-f2b548775f14
- Over MCP: `get_schema_card` with id `97a98fc5-e6f9-555c-a22f-f2b548775f14` at https://api.scigantic.com/mcp

## Guide

- [How to query ChEMBL bioactivity data as Parquet without installing the database](https://scigantic.com/guides/open-chembl-bioactivity-data)

## Related topics

- [Drug Discovery datasets](https://scigantic.com/topics/drug-discovery)
- [HDF5 datasets](https://scigantic.com/topics/hdf5)
- [FASTA datasets](https://scigantic.com/topics/fasta)
