Computed or measured?

A data sheet staples two kinds of information together. Some fields were computed from the drawn structure by a program; others were measured by someone, somewhere, and copied from a paper. Telling the two apart explains why some numbers agree across every source to the last digit and others do not.

Derived from the structure

• Formula, average mass, monoisotopic mass: counted and summed from the structure and the atomic weights. Deterministic. Two sources disagree only if they use atomic weights from different years, and then only in the third decimal of the average mass.
• Identifiers such as SMILES and InChI: generated from the connectivity and the wedges, so the same structure always yields the same string.
• Predicted properties: a calculated partition coefficient, a count of hydrogen-bond donors, a polar surface area. Model outputs, consistent within one source and not necessarily between sources.

Measured in a laboratory

• Melting and boiling points, solubility, density: taken from the literature, with the citation on the record. Values vary with purity and method, so ranges are common.
• Measured mass: a mass spectrum confirms the monoisotopic mass to within a few parts per million. Agreement between the computed and the measured value is how a formula is assigned to an unknown.
• Spectra (NMR, IR): experimental evidence for the connectivity the drawing claims.

How to read a record

Treat computed fields as restatements of the structure: if the drawing is right, they are right. Treat measured fields as claims with a provenance, and follow the citation when the number matters. Well-run resources mark the difference: ChEBI lists its atomic weights and the display precision it keeps; PubChem labels each property as computed or experimental and names the source of each. Two masses, one molecule covers the two computed masses a record carries, and What a mole actually counts the units they are quoted in.