How the compound library is built

Each entry in the compound library answers six questions about one substance: what it is, what its molecular identity is, what has been established about its chemistry and molecular interactions, which experiments established it, what remains uncertain, and what to read first. This page explains how those answers are reached and what the library will not say.

What each entry is based on

Entries are written from primary literature and authoritative records: the original discovery papers, structure papers, PubMed and PubMed Central, PubChem, ChEBI, UniProt, the RCSB Protein Data Bank, the IUPHAR/BPS guide, nonproprietary-name lists and patents where a modification is proprietary. The target is at least 20 distinct relevant works per entry, read substantively, with at least 8 primary studies where the literature has them.

Counting is strict. An abstract, its PubMed Central copy and the publisher's version are one work. A search snippet, a failed fetch or an off-topic page counts as nothing. Where only the abstract was read, the reference says “abstract only”; a full text is never claimed for a paper that was not opened. Where fewer than 20 credible works exist, the entry says so and names what was searched. Sparse literature is reported as a finding, not padded. Every reference is checked for retraction, correction and expressions of concern, and the status is printed beside it.

How identity is decided

Identity comes first, before anything is said about what a compound does. Sequence, termini, D- and L-residues, non-standard residues, cyclisation, disulfides, lipidation, linkers, metal stoichiometry and salt form are each checked against the records. A product label is treated as a claim, never as proof.

Each entry carries one of three identity states. Verified means the public record pins the name to one chemical species. Partially verified means the name could denote more than one distinct substance, and the entry names what is open. Unresolved means no candidate can be preferred on the record; the entry then lists what the material could be and the exact question the records could not answer. A value that is unknown is shown as unknown, with the reason. No sequence, registry number, reference or figure is ever filled in by inference.

A blend or formulation is described by its components. A mixture has no single formula, mass or structure, and a study of one component is not a study of the blend, so the library never infers synergy, compatibility or an additive effect.

How the molecular picture is chosen

The 3D view on each page comes from the strongest source that genuinely fits, in this order:

1. Experimental structure: a deposited experimental structure of the exact molecule.
2. Experimental, in a complex: the exact molecule inside a deposited complex, such as a receptor or G-protein complex. The compound's own chain is named, its sequence is checked residue by residue against the verified one, and anything not modelled is listed.
3. Calculated conformer: a computed 3D conformer from PubChem for the exact record, labelled as calculated.
4. No 3D model: none of the above, or a mixture, a heterogeneous preparation, a flexible peptide with no deposited structure, or an unresolved identity. The page says why.

A computed model is never presented as experimental, and no molecule is ever shown because its name resembles the one the page is about. Every structure file is stored with its source, retrieval date and checksum, and the checks it passed are printed under the viewer.

Figures

Figures from third parties are used only under a licence read from the host's own records, and each is printed with its credit, licence and a link to its source. Figures made for the library are schematics: boxes and arrows standing for something the text has already argued. They never depict a molecule, because an invented chemical drawing would be a confident wrong picture on a page whose purpose is to be checkable.

What automated review checks

Every entry passes an automated review, run separately from the one that wrote it, before it can be approved. It fetches every identity figure and cited record again from the primary databases rather than trusting the author's notes, checks every cited paper for retraction, spot-checks citations against the papers they cite, and reads the text against the rules below. Defects it finds are fixed before the entry passes.

The review is automated and is named as automated on every page. No human scientific review is claimed for any entry. An entry is published only after the owner has approved it.

What the library will not say

The library describes chemistry, structure and published laboratory findings. Findings are stated as a measured endpoint in a named experimental system, with what the experiment did not establish beside it. A result in one system is not evidence about another, and the pages say so.

It contains no guidance on use of any kind: no doses, routes, schedules, preparation for use or combinations, no claims of benefit, and no prices or purchasing information in any entry's scientific content. Clinical trials may be cited as bibliography where they matter historically; their regimens and patient outcomes are not reproduced.

Dates and corrections

The “Updated” date on an entry is the date its research or its automated review last changed. It does not move when the site is rebuilt. A reference that has been corrected, retracted or given an expression of concern is marked as such in the entry's reference list.

To report an error in an entry, use the contact page and name the entry and the claim.