SHEET /about/methodology · METHODOLOGY

Methodology

MoleculeFinder is assembled from public-domain and openly licensed data, then labeled and cross-linked so every claim is traceable. Nothing is scraped from a source we are not free to use, and the whole site rebuilds from a committed snapshot, so the same inputs always produce the same pages.

Where the data comes from

Each field is attributed to one of a small, fixed set of sources:

• PubChem (public domain): molecular identity, structure (SMILES and InChI), physical properties, synonyms, reported animal toxicity (LD50), and GHS hazard classifications.
• Wikidata (CC0): the preferred name and a short factual description. We use only the CC0 description text, never Wikipedia prose.
• Wikimedia pageviews: how often a molecule is looked up, used only to rank which molecules to feature and build first.
• RDKit (BSD): everything drawn or calculated from a structure, described below.
• MoleculeFinder curated: hand-authored facts for the marquee molecules. The foods a molecule is found in, its short editorial, its half-life, sweetness, pungency, and type are written and checked by us.

What we compute

From each molecule's structure we compute its 2D depiction, standard physicochemical descriptors, and its nearest structural neighbors (Morgan fingerprints compared by Tanimoto similarity). The interactive tools run transparent, unit-tested formulas over stored inputs plus what you tell them: a first-order decay curve for "still in your system", an LD50 scaled to body weight for the dose lens, the Widmark equation for blood alcohol, and the Scoville scale for pungency. Every computed value is labeled as computed.

The license firewall

We never ingest a license-restricted database. Food-composition, flavor, and odor sets that forbid commercial reuse are off limits, so the foods and sensory notes you see are hand-curated rather than lifted. This keeps the compilation free to publish and free to use.

How the canon is chosen

The set of molecules is led by demand: the most-looked-up molecules come first, alongside a hand-picked marquee tier of the ones behind everyday life. The library grows in stages, the quality core first, so pages stay rich and cross-linked rather than thin.

Reproducible by design

The site is static, built from a versioned data snapshot with no live database on the page. The pipeline that produces it is idempotent: re-running it reproduces the same records, and the snapshot's history is the changelog.

See how every value is labeled, or contact us.