# Approved covalent drugs in ChEMBL 37

Data behind the ProteinIQ guide [How many covalent drugs are approved?](https://proteiniq.io/guides/approved-covalent-drugs).

Last checked: October 29, 2026. We plan to rerun the analysis after each new ChEMBL release.

## Files

- `covalent-drugs.csv`: the 173 approved small-molecule drugs that meet the broad definition, one row per drug, with ChEMBL ID, name, first approval year, warhead patterns that matched, ChEMBL's mechanism text, the hand-review decision, `definition` (`strict` or `broad`), warhead class, bond type, reversibility, target class, target, evidence grade and the basis for the decision.
- `covalent-drug-candidates.csv`: all 228 candidates (every drug that matched a warhead pattern, a mechanism term or our named list), including the 55 that were reviewed and rejected, with the reason.
- `covalent-drugs-review.tsv`: the hand-review decisions as made, the input to the two files above.
- `covalent-drugs-outside-chembl.tsv`: covalent drugs that the ChEMBL 37 approved small-molecule set does not contain (approved after its 2024 cut-off, approved only in China, or dropped by the population rules), checked by hand. These are not part of the main counts.
- `summary.json`: every aggregate quoted in the guide, including counts by warhead, target class and decade.
- `analyze-covalent-drugs.py`: the script that produced these files. It is a copy of `scripts/covalent-drugs-chembl.py` in the ProteinIQ repository.

## Population and method

The population is the 2,104 approved small-molecule drugs of the ProteinIQ [Lipinski analysis](https://proteiniq.io/guides/lipinski-rule-of-five): ChEMBL release 37 (released May 1, 2026) molecules with `max_phase` 4, collapsed to parent compounds, small molecules and defined peptides up to 900 Da, without metals or several components. Structures and mechanisms were read from the ChEMBL REST API on October 29, 2026.

Each structure was searched with RDKit 2025.03.5 for 34 warhead substructure patterns. ChEMBL mechanism records were searched for the terms covalent, irreversible, alkylating, acylating, cross-linking, suicide, mechanism-based, thiol and cysteine. A list of drugs known to act covalently through groups the patterns cannot see (for example proton pump inhibitors) was added. The 228 candidates were reviewed by hand against drug labels, ChEMBL mechanisms and the literature.

| Term | Definition |
|---|---|
| Broad | The drug, or an active metabolite formed in the body, forms a covalent bond with its therapeutic target (protein, DNA or RNA) as part of the accepted mechanism, reversible or not. Adducts with freely dissociating coenzymes (NAD, NADP, PLP) do not count. |
| Strict | A broad drug whose bond to a defined residue of a protein is the designed basis of activity: a recognition scaffold plus a reactive group installed or tuned against that target, confirmed by structure or mass spectrometry. Drugs found by phenotypic screening, by chance or as natural products before the mechanism was understood stay in broad only. |
| Evidence A | The adduct with the target is shown by crystal structure, mass spectrometry or the drug label. |
| Evidence B | The covalent mechanism is accepted, but the adduct is not directly characterized or the residue is not assigned. |

## Reproduce

```bash
python scripts/covalent-drugs-chembl.py --cache-dir /tmp/covalent-drugs
```

The script needs the Lipinski output `public/data/guides/lipinski-rule-of-five/approved-drugs-ro5.csv` and the review file `scripts/covalent-drugs-review.tsv`. It stops if a candidate has no review decision. Delete the cache directory to read a new ChEMBL release.

## License

ChEMBL is licensed under [CC BY-SA 3.0](https://creativecommons.org/licenses/by-sa/3.0/), so these files are shared under the same license. Please credit ChEMBL (Zdrazil et al., Nucleic Acids Research, 2024) and ProteinIQ, with a link to https://proteiniq.io/guides/approved-covalent-drugs.