One molecular formula, 217 materials under it, from rose to mushroom

About a 4 minute read.

Too long, didn't read

• 21,905 materials carry a formula. Drop the 295 that say unspecified and 21,610 spread across 6,710 distinct formulas
• 2,181 of those formulas are shared by two or more materials, covering 17,081 materials, or 79%
• C10H18O holds 217 materials; 78 carry odor data, spread across 19 odor families
• All 217 weigh 154.25. Longevity across them runs from 1 hour to 388
• Search on a formula and you get a shelf, not a material. For uniqueness, use the CAS number

A search that arrived as a formula

Last month's Search Console holds a query reading c₁₀h₁₈o, subscript digits and all. Five impressions, no clicks. Somebody is looking for materials by formula.

That query sent me to check something I had never checked: in this database, how many materials does one formula carry on average?

79% of materials do not have their formula to themselves

21,610 usable formula records spread across 6,710 distinct formulas. That averages 3.2 materials per formula.

The average is useless here, because the distribution is lopsided. 2,181 formulas are shared by two or more materials, and those formulas cover 17,081 materials. Pick a material with a formula at random and there is a 79% chance somebody else is standing under the same formula.

The ten most crowded:

Formula	Materials	Roughly what lives there
C15H24	293	Sesquiterpene hydrocarbons, the cedar and patchouli skeletons
C10H18O	217	Monoterpene alcohols and ethers, the subject of this piece
C10H16O	210	Monoterpene ketones and aldehydes
C15H24O	204	Sesquiterpene alcohols
C15H26O	175	Sesquiterpene alcohols, one degree more saturated
C10H18O2	171	Monoterpene diols and esters
C12H22O2	141	Medium-chain esters and lactones
C11H20O2	123	The same, one carbon shorter
C8H14O2	105	Short-chain esters

The sesquiterpene hydrocarbon square holds 293. Search C15H24 and you get close to three hundred answers.

Who is standing in the C10H18O square

Of the 217, 78 carry an odor field, and those 78 fall into 19 odor families:

Odor family	Count
floral	14
herbal	9
minty	9
citrus	8
fatty	6
woody	6
balsamic	4
camphoreous	4
11 more families, 1–3 each	18

Sorted by supplier count, the top of the square looks like this:

Material	CAS	Suppliers	Odor family	First descriptors	Longevity (hr)
Geraniol	106-24-1	198	floral	sweet, floral, fruity, rose	60
Linalool	78-70-6	135	floral	citrus, floral, sweet, bois de rose	12
alpha-Terpineol	98-55-5	103	terpenic	pine, terpenic, lilac	20
Nerol	106-25-2	85	floral	sweet, neroli, citrus, magnolia	44
1,8-Cineole	470-82-6	77	herbal	eucalyptus, herbal, camphoreous, medicinal	4
Citronellal	106-23-0	70	floral	sweet, dry, floral, herbal	16
(±)-Menthone	89-80-5	42	minty	minty	12
(E)-2-Decenal	3913-81-3	27	fatty	waxy, fatty, earthy, green, cilantro, mushroom	57

Rose, eucalyptus, mint, mushroom. One formula.

They also all weigh 154.25, because molecular weight is computed from the formula and carries no information the formula does not already have. Longevity is where they split: 47 of the 217 have a longevity record, running from 1 hour to 388, a 388-fold spread.

The difference is in how the skeleton connects, not how many atoms there are

A formula states a list of atoms: ten carbons, eighteen hydrogens, one oxygen. It says nothing about how they connect. The same list splits along at least three routes.

Different skeleton. 1,8-Cineole is a bicyclic ether, geraniol an open-chain alcohol. One smells of eucalyptus lozenges, the other of rose.

Same connections, different orientation. Geraniol and nerol are the cis/trans pair of the same chain. The database gives them 198 and 85 suppliers, 60 and 44 hours of longevity. Both odor fields open with sweet, but one goes towards rose and the other towards neroli and magnolia.

Same orientation, mirrored. That is what the paper in the next section examined.

Worth noting that the database itself files the mirror images separately. linalool (135 suppliers, 12 hours) and laevo-linalool (25 suppliers, 16 hours) are two records with different descriptors. The first reads citrus, floral, bois de rose, blueberry; the second reads fresh, floral, woody, natural, lavender. Catalogue data treats the mirror images as two products.

The Hiroshima experiment: mirrored linalools differed down to the brainwaves

In 2000, Sugawara and colleagues at Hiroshima Prefectural Women's University published a careful piece of work in Chemical Senses (PMID 10667997). Using repeated flash column chromatography, they isolated (R)-(−)-linalool (specific rotation −15.1°) from lavender oil and (S)-(+)-linalool (+17.4°) from coriander oil, with commercial material as the racemic (RS) form (0°, 50.9% R and 49.1% S).

Subjects inhaled each of the three through an inhalator, once before and once after work, with two kinds of work: listening to environmental sounds, and mental arithmetic. Two things were measured, a subjective sensory questionnaire and forehead surface EEG (IBVA-EEG).

The result, in the paper's own terms: inhalation after work evoked different subjective impressions depending on the configuration of the isomer and the type of work. After listening to environmental sounds, (R)-(−)-linalool produced a much more favourable impression on the questionnaire, accompanied by a greater decrease in beta waves relative to before work. After mental work, the same material tended towards agitation, with beta waves rising.

The authors concluded that the enantiomeric stereospecificity of linalool evoked different odor perception and responses, both chirally dependent and task dependent.

Read it carefully. The sample is small, what is measured is a forehead surface potential rather than receptor activity, and the effect itself is task dependent, with the same material moving in opposite directions depending on what came before. One conclusion holds: two mirror images are not the same olfactory stimulus. The list of atoms is identical.

The model in Science does not eat formulas either

The second paper is the principal odor map (POM) in Science in 2023 (Lee et al., PMID 37651511, a collaboration between Google Research and the Monell Chemical Senses Center). They used graph neural networks to place molecules on a map that preserves perceptual relationships, then used it to predict odor quality for molecules nobody had described.

The score is worth copying out. On a prospective validation set of 400 out-of-sample odorants, the model-generated odor profile matched the trained panel mean more closely than the median panelist did. Its reliability in putting a smell into words reached the level of a trained human panelist.

The disclosure belongs here as the paper states it: the work was funded by Google Research, several authors joined Osmo Labs during the review process and hold ownership interests, and the competing-interest statement runs long.

What I am pointing at is not the accuracy though, it is what goes in. A graph neural network receives the molecular graph: which atom bonds to which, by what bond, in what stereochemistry. A formula is what is left of that graph once every edge has been thrown away. Not one of the best current structure-odour models starts from a formula.

So what should you search on

• For uniqueness, use the CAS number. C10H18O returns 217 candidates; CAS 106-24-1 returns the geraniol record. This site's search accepts names and CAS numbers.
• If a formula is all you have, walk it as a shelf. Knowing you are receiving 217 rather than 1 is itself worth having.
• The same CAS is no guarantee of uniqueness either. Same CAS, different odor family measured that one. CAS is stricter than a formula, but it is not perfectly clean.

→ Search by odour: all 42,225 materials, searchable by odour, name or CAS.

What this article does not establish

• Only 78 of the 217 carry an odor field. The other 139 have no data, so the claim is not that 217 materials all smell different, only that the 78 known ones spread across 19 odor families.
• All of them weighing the same is a definition, not a finding. Molecular weight is computed from the formula; that row is here to show how little a formula carries.
• Sugawara's sample was small and the measurement was forehead EEG, not receptor activity, and the effect shifted with the task.
• The POM predicts odor descriptors, not longevity, strength or safety. Those fields still need measuring.
• The odor family labels come from the source data, not from any relabelling on our part; the number 19 carries the granularity of the original annotation.