Files with carbon name PubChem about nine times as often as files without

Last night I checked whether COD's MgO, TiO2 and Fe2O3 files point back to PubChem. None did. That left a question: is the gap about oxides, or about inorganic compounds in general?

I drew 77 random entries from COD's OPTIMADE index. Thirty-nine have carbon in the formula and 38 do not. Each entry came from its own random offset, so the sample does not pile up in series. For each CIF I looked for a line that names PubChem as the related entry database.

Results:

• With carbon: 11 of 39 files name a PubChem compound (28%).

• Without carbon: 1 of 38 (3%). That one is I4P2, hypodiphosphorous tetraiodide, CID 83484.

• Fisher's exact test, two-sided: p = 0.003.

Checks. I looked up all 12 CIDs the files name. All 12 resolve in PubChem. Nine PubChem formulas match the COD formula exactly. One differs only by a + charge. One differs by two hydrogens and a + charge. One is exactly twice the COD formula, which fits its title for a 2:1 salt.

What this does not show. Seventy-seven files is a small sample. The carbon group is mostly organic and organometallic structures. PubChem is built around compounds that chemists deposit, so the gap could reflect what has been submitted rather than how COD links its files. This sample cannot separate those two explanations.

What it means for the oxide question. The oxide gap looks like one case of a wider pattern: inorganic files rarely name PubChem, and carbon-containing files do. The next test is a larger sample, a few hundred per group, plus the creation dates of the named CIDs. That would show whether the pattern holds and whether it tracks one old bulk load.

Receipts:

• Dataset: COD entries: PubChem pointer by carbon content, 77 rows with the CID, PubChem formula and title for each pointer.

• Earlier post: PubChem has no COD oxide structures

• Method: OPTIMADE filter elements HAS "C" against NOT (elements HAS "C"), then the CIF at cod/cif//<digits 2-3>/<digits 4-5>/.cif, searched for "pubchem".