1 June 2019 | New Scientist | 45hunter-gatherer ancestors evaluated the
freshness of potential foods. “The larger the
molecule, the more intact the material you’re
smelling is,” says Keller. “Decay bacteria eat
the big compounds and break them down into
smaller molecules. The smaller the molecules,
the further away you are from life.”
Artificial intelligence can go even further
in predicting odours. In 2017, Keller and his
colleagues announced the results of their
DREAM Olfaction Prediction Challenge, an
international competition to see if there was
a way to tell how a molecule smells just from
its structure. Competitors were given data on
more than 400 different molecules rated
according to their pleasantness and other
qualities such as fruitiness, spiciness and
floralness, as well as physical descriptions
of each molecule. The aim was to find out
which machine-learning algorithm was
best at predicting sensory attributes from
molecular structure. Once entries were in,
the researchers tested the accuracy of each
program’s predictions on 69 other molecules.
The winning algorithm, devised by Rick
Gerkin at Arizona State University and his
colleagues, was roughly 50 per cent accurate
in predicting pleasantness ratings. That is
better than it sounds, because a person’s
rating of the same molecule varies somewhat
from moment to moment. That variation sets
an upper limit for accuracy – in this case, at
about 65 per cent – for any algorithm. Gerkin’s
algorithm had, in other words, predicted
people’s pleasantness ratings almost as well
as knowing how they had rated an odour on a
previous encounter. It could also successfully
predict descriptors such as “garlic”, “fish”,
“burnt” and “sour”, raising the possibility
of predicting the qualities of almost any
molecule, and even of reverse engineering
odorants by starting with a target smell and
then making a molecule with that odour.
Whose nose
But these efforts have a big shortcoming: they
focus on individual molecules. “There is no
situation where you encounter just one type
of odour molecule in nature, so all of us have
been studying a stimulus that the brain didn’t
evolve to decode,” says Sobel. “It would be a bit
like saying you’ll try to understand how the
brain processes language by studying pure
tones. It just doesn’t work that way.” Sobel
is now investigating how people perceive
mixtures of odorants, though he won’t say
much about his latest results because they
are yet to be published.
In the meantime, we are left with expert
hunches about what makes a pleasing mix
of odours. “My best guess, which is really not
very good, is that we love complexity,” says
Luca Turin, a biophysicist at the Alexander
Fleming Biological Sciences Research Centre
in Vari, Greece. “We like an orchestral sound
in smell space, so to speak.” The hundreds of
chemicals in roasted, toasted compounds
such as coffee or bread give a richness to their
aromas that simpler mixtures can’t match.
Master perfumers have a few tricks of their
own when it comes to creating that richness
in their own concoctions. A typical fragrance –
whether intended for soap or a fine perfume –
will usually have several dozen ingredients,
and the perfumer must get the dose right for
each one so that no single smell dominates.
Perfumers also layer ingredients that
evaporate at different rates, so that the
fragrance evolves over time from evanescent
top notes towards more lingering base notes.
Surprisingly, it matters very little whether
individual ingredients smell pleasant. To add
complexity to their formulas, perfumers
often include small amounts of unpleasant
sounding “animalic” ingredients, such as
faecal-smelling skatole or the anal secretions
of the civet, a small mammal from Asia and
Africa (see “Scents from strange places”, left).
The desire to spice up scents is driving
the hunt for new molecules for fragrance
companies to add to their formulas. That is
what leads Piquart to the Peruvian Amazon
and many other remote parts of the world.
There is no shortage of new ingredients to
discover, he says. “There are many, many
things that smell, many possibilities when I
go to a country,” he says. “We have discovered
only 20 per cent, maybe.”
Many companies, though, prefer to have
their chemists synthesise the key odorants in
the lab (see “Secrets of a fragrance laboratory”,
page 42). With this approach, who knows
what exotic molecules we will be smelling
in the future.
Where does all this leave us in the hunt
for what exactly makes a good smell? The
perception of smells is a complex, subjective
and counter-intuitive business, where even
seemingly disgusting odours can be attractive.
There is no simple formula. But that isn’t
something to turn your nose up at. ❚“ Surprisingly,
it matters very
little whether
individual
ingredients
smell pleasant”
When it comes to prized
perfume ingredients – now
made synthetically, you’ll be
pleased to hear – our tastes
are far from vanilla.Ambergris
This rare aroma from intestinal
secretions of sperm whales,
found washed up on beaches,
is one of the most expensive
natural substances. It has been
used in perfumes, medicines
and aphrodisiacs for centuries.Skatole
Named after skato, the ancient
Greek word for excrement, this
compound is found in faeces,
but also in some flowers. Foul-
smelling at high doses, it has a
flowery aroma at lower levels.Castoreum
Produced as a scent marker,
this compound from glands
near a beaver’s anus has
musky, fruity notes.Civetone
This strong, musky secretion
from the anal glands of civets,
smells pleasant at low
concentrations.SCENTS FROM
STRANGE PLACESBob Holmes is a consultant for New
Scientist and author of Flavour: A
user’s guide to our most neglected
sense (WH Allen)