How PepBank works

Studies come from Europe PMC, with publication types from PubMed. Trials come from ClinicalTrials.gov, and retractions from Retraction Watch. Peptide details come from UniProt, PubChem, GSRS and ChEMBL. Approval and sport status come from the FDA, EMA and WADA. The source list gives the license of each one.

1. Read. An AI model (Claude Sonnet) reads the abstract of a study, the short summary at the top of the paper. It writes down each result and copies the exact sentence that the result comes from. If the abstract gives no number, the result has no number.
2. Check. Software confirms that every quote and every number appears in the abstract. A second AI model then reads the abstract again and accepts, fixes or rejects each result.
3. Publish. Results that pass both checks appear on peptide pages, in search and in the API. Each one links to its source sentence.

AI wrote and checked these results. AI can make mistakes, so check important results against the source.

The label says what kind of study it was. It does not grade the study. A result in animals or cells does not show the same effect in people.

Some studies do not give the peptide. They measure the peptide that the body makes, or they use the peptide as a tool to test something else. These results appear in their own sections, and they do not count as evidence for taking the peptide.

The key results at the top of a peptide page follow a fixed order: studies in people first, then stronger study designs, then larger, more cited and newer studies. One study cannot fill every place.

A retracted study stays on PepBank with a warning, and its results never count as evidence. PepBank compares every study with the Retraction Watch list. Today it marks 133 retracted studies, 20 expressions of concern and 1,019 corrected studies.