Beginner perfumer · 118
Beginner perfumer #118: of 42,225 materials, about 4,850 are fragrance materials
· 18 min read
This database's front page says 42,225 materials. This round I cut it four times. First cut: 6,457 records whose category field itself reads 'information only not used for fragrances or flavors'. Second cut: 2,829 blends whose names contain fragrance, specialty, flavor or replacer, a group where only 3% carry a CAS number against 79% of everything else. The third cut is the largest: 27,277 records have an entirely empty odour field, 65% of the database, because it also holds cosmetic ingredients, food additives and pharmaceuticals. Fourth cut: 808 records with no suppliers at all. What remains is 4,854 records, 11.5%, that have suppliers, have an odour, and are not marked as non-fragrance. And that number explains several of my earlier misreadings in retrospect: a logP range sweep flagged 320 anomalies of which 319 were correct, and a vapour-pressure regression's largest outliers were acyclovir and hydrocortisone.
About a 5 minute read.
The short version
| Cut | Removed | Remaining |
|---|---|---|
| Start | 42,225 | |
| ① Category field says "not used for fragrances or flavors" | −6,457 | 35,768 |
| ② Name says it is a blend or a replacer | −2,829 | 32,939 |
| ③ Odour field entirely empty | −27,277 | 5,662 |
| ④ No suppliers at all | −808 | 4,854 |
- 4,854 records remain (11.5%) with suppliers, an odour, and no non-fragrance marking
- Requiring a CAS number too: 4,621 (10.9%)
- The largest cut is the empty odour field, at 65% of the database
- Only 3% of the blend records carry a CAS number, against 79% of everything else
Why cut at all
Last round's sweep of specification ranges turned up
something: half the wide-range records had fragrance or specialty in the name. Those are blends,
and they cannot have a single specification in the first place.
That prompted a more basic question: of these 42,225, how many are things I could buy a bottle of and smell?
First cut: the database says so itself
The category field has one value that reads:
information only not used for fragrances or flavors
The database's own words.
That label sits on 6,457 materials. The best-supplied of them:
| Suppliers | Material |
|---|---|
| 21 | iodine |
| 20 | musk xylol |
| 20 | geranyl nitrile |
| 18 | musk ambrette |
| 15 | sodium cyclamate |
| 13 | lactose monohydrate |
Iodine, lactose and cyclamate are not fragrance materials. Musk ambrette and musk xylol are things that used to be.
Second cut: the name says so
2,829 records have fragrance, specialty, flavor, replacer, accord or reconstitution
in the name.
The difference between that group and everything else is very clean:
| Blend records | Everything else | |
|---|---|---|
| Has CAS | 3% | 79% |
| Has molecular formula | 1% | 56% |
| Has suppliers | 98% | 47% |
No CAS number, and almost always suppliers. That is exactly what a blend should look like: it is not a compound, so there is no registry number, and every company makes one, so there are plenty of suppliers.
The best-supplied entries say it all:
| Suppliers | Material | Category field |
|---|---|---|
| 84 | vanilla flavor | flavored products of all types |
| 78 | strawberry flavor | flavored products of all types |
| 73 | orange flavor | flavored products of all types |
| 71 | rose specialty | fragrance agents |
Eighty-four suppliers sell "vanilla flavor" because each of them makes one. That record is a shelf label, not a product.
Third cut: the odour field is empty
This one removes 27,277 records, 65% of the database.
The reason is simple: this database does not only hold fragrance materials. It also holds cosmetic ingredients, food additives, pesticides and pharmaceuticals. Nobody writes an odour field for those, because nobody smells them.
This explains several of my earlier misreadings
That cut let me understand a few things in retrospect:
- When I swept on a logP range I flagged 320 records and 319 of them were correct, because I applied a fragrance chemist's intuition to a database full of cosmetic ingredients
- When I caught boiling-point anomalies with a vapour-pressure regression, the largest residuals were acyclovir, spironolactone and hydrocortisone plus a pile of herbicides; only 4 of the 89 positive residuals were fragrance materials
- When I measured the usable aliases in the
synonymsfield, the "usable" ones weresilver,tibetanandafrican: adjectives from botanical common names
Three times, the same cause. I thought I was sweeping a fragrance database.
Fourth cut: no suppliers
Another 808 go: they have an odour, are not blends, are not marked non-fragrance, and nobody sells them.
Most are constituents that appeared in the literature, or things that went out of production. Longifolene's neighbours include several like this.
What the remaining 4,854 look like
To check the cuts landed right, here are the best-supplied survivors:
| Suppliers | Material | Odour type |
|---|---|---|
| 198 | geraniol | floral |
| 193 | vanillin | vanilla |
| 148 | patchouli oil | woody |
| 135 | citral | citrus |
| 135 | linalool | floral |
| 130 | geranyl acetate | floral |
| 129 | eucalyptus globulus oil | herbal |
| 128 | clove bud oil | spicy |
That is what sits on a perfumer's bench. The cuts are right.
Incidentally, the replacers tell a story
Among the 2,829 the second cut removed, one class stands out: replacer.
The 954-formula corpus holds 24 replacers across 178 entries. Set them beside the originals:
| Original | Corpus count | Replacer | Corpus count |
|---|---|---|---|
| musk ambrette | 0 | musk ambrette replacer | 56 |
| rose absolute | 0 | rose absolute replacer | 9 |
| bergamot oil | 17 | bergamot oil replacer | 28 |
| musk xylol | 2 | musk xylol replacer | 13 |
| jasmin absolute | 23 | jasmin absolute replacer | 13 |
| neroli oil | 17 | neroli oil replacer | 10 |
| civet absolute | 11 | civet absolute replacer | 5 |
The two originals scoring zero are exactly the two you cannot get. Musk ambrette is banned
(its category field says not used for fragrances, its regulatory field says FEMA GRAS (withdrawn)),
and rose absolute costs too much to appear in a demonstration formula.
And bergamot's replacer outnumbers the original.
How often a replacer is used measures how hard the original is to obtain.
How to use this
When you look something up in this database:
- Check the category field first. If it reads
information only not used for fragrances or flavors, stop reading - Check the name for
flavor,fragranceorspecialty. If it is there, that is a shelf label - Check whether the odour field is empty. If it is, nobody treats this as an odour material
- Check for a CAS number. No CAS with suppliers usually means a blend
Four rules, none requiring any chemistry. You only look at whether fields are there.
What this doesn't establish
- The four cuts are mine, and their order changes the result. Removing empty odour fields first would shrink all three other numbers. I report them applied in sequence.
- 11.5% does not mean "only 11.5% of this database is useful". Those 27,277 empty-odour records are useful to people working on cosmetics or food; they are just not useful for perfumery.
- "The odour field is empty" is not "it has no odour". It may simply be unfilled.
- The second cut judges by keyword. I tried a field rule instead (no CAS, has suppliers) and it
catches 4,503, but it wrongly picks up real materials such as
rose hips seed oilandcardamom seed oil CO2 extract. Those are natural extracts without a CAS number, not blends. So I settled on the name. - The replacer table shows eight pairs. The corpus holds 24 replacers; I picked the ones whose original also appears in the corpus.
- The 4,854 has not been re-checked for quality. Longifolene is in this group with five unusable fields. Passing these four cuts means it is a fragrance material, not that its data is right.