Attempts at Olfactory Space Shaping

In bits and pieces, we are seeing the organization of olfactory precepts become a possibility. It's really only been happening in the past five years or so, and because of a couple changes in the game, namely better and newer data, and better algorithms to do the prediction. And yes, those better algorithms are powered by the new-style machine learning artificial intelligence that is blasted from every headline on the tech feed -- deep learning. 

I posted something not long ago about the 2015 DREAM Olfaction Prediction Challenge, and the new dataset that made it all possible.

Today we're looking at another group of ambitious osmologists who figured out some good rules for organizing the category-averse sensory world of smells. Their results are quite different from those of the DREAM Challenge, so I thought it was worth it to summarize their results. 

Their database comes from the Dragon set of chemoinformatics (odorous molecules and their chemical properties) and the Arctander set of olfactory descriptors (chemicals and the names they are likely to be associated with). I'm not sure why they chose the Arctander set instead of the newer Keller Vosshall set. They reviewed other sets and pretty extensively, such as the Dravnieks set, FlavorNet, and one other, but they don't mention the Keller Vosshall set, which is the newest of the bunch, so maybe they're waiting until it proves its worthiness.

The Data

Their database ends up with 1689 molecules each with 82 chemical attributes, and corresponding with 74 olfactory descriptions, which they use to generate "rules" for matching the chemical info to the olfactory info. They call it "more of an exploratory data analysis than an accurate prediction machine." But that's ok, because the science of predicting odor qualities by their chemical attributes is still in its exploratory phase.

The Rules

They correlate their chemicals to odor-names finding the combinations of chemical conditions that produce subsets of odor categories, and came up with 473 physiochemical rules.

The Results

First, we have some typical odor categories and the chemical features they were found to associate with:

Floral - either aromatic and strongly hydrophobic molecules or non-aromatic and moderately hydrophobic odorants.

Camphor - molecules are rather small in size, moderately hydrophobic, and eventually cyclic.

Earthy - moderately hydrophobic molecules with unsaturations.

Spicy - rigid molecules, eventually aromatic.

Woody - hydrophobic molecules, rather not cyclic nor aromatic.

Fatty - larger carbon-chain skeleton which is highly hydropobic with aldehyde or acid functions.

Fruity - moderate hydrophobicity and being medium to large in size.

***

And here is their list of subsets that do a good job of organizing all the 1600 molecules:

Sulfuraceous - encompass molecules with one or two sulfur atoms and are moderately heavy, with a maximum of six carbon atoms.

Phenolic - moderate size, with few unsaturations and low hydrophilicity (and high lipophilicity). It can be regarded as a cyclic molecule.

Vanillin - mostly cyclic molecule (like the prototypical molecule vanillin), with 3 Hydrogen bond acceptors branched on saturated carbons atoms on an aromatic cycle.

Musk - heavy and hydrophobic compounds. This is reflected by a rather large logP, surface area or molecular weight.

Sandalwood - (A diverse set; minor modifications within their structure can abolish the sandalwood note. The rules which are mined here correspond to models which are very simple and hardly capture the subtlety of this odorant family.)

Almond - at least one oxygen and/or other hydrogen bond-accepting atom but also bearing an aromatic cycle. This means that the structure bears several unsaturations. These chemicals are thus relatively small. [benzaldehyde is representative]

Orange-blossom - diverse structures ranging from very small to medium or large compounds. As a general rule, one can note the presence of unsaturations, consistent with a terpenic structure, associated with a quite hydrophobic feature.

Jasmine - (i) molecules composed mainly of carbons and oxygen atoms, (ii) molecules with an aromatic core and embranchments conferring a large flexibility, and (iii) compounds with an optimal chain length around five carbon atoms. [jasmonate is representative]

Hay - hydrophobic molecules composed of aromatic cycles, being either heterocyclic or linked to a heteroatom outside of the cycle. These atoms confer to the molecule the possibility to accept Hydrogen bonds.

Tarry - (not easy to establish specific characteristics of the molecules of this group, but overall these molecules are flexible, presenting heteroatoms while having low hydrophilicity due to the presence of double bonds.)

Smoky - (a robust rule is hard to establish because the physicochemical descriptors refer either to aromatic compounds with a hydroxyl group or flexible molecules with rotatable bonds.)

In Conclusion

This explains why I wrote a book about the language of smell:

"The more one moves towards the area of perceptual space of odors that is characterized by its heterogeneity between individuals, the higher the predictability threshold (i.e. bad prediction) becomes. This variability characterizes what could be called "the glass ceiling of olfactory diversity".

image source link

Notes:

Chemical features mining provides new descriptive structure-odor relationships

Carmen C. Licon, Guillaume Bosc, Mohammed Sabri, Marylou Mantel, Arnaud Fournel, Caroline Bushdid, Jerome Golebiowski, Celine Robardet, Marc Plantevit, Mehdi Kaytoue, Moustafa Bensafi. PLoS Comput Biol. 2019 Apr; 15(4): e1006945. Published online 2019 Apr 25.

doi: 10.1371/journal.pcbi.1006945

Post Script:

Pleasantness and trigeminal sensations as salient dimensions in organizing the semantic and physiological spaces of odors. C. C. Licon, C. Manesse, M. Dantec, A. Fournel, and M. Bensafi. Sci Rep. 2018; 8: 8444. doi: 10.1038/s41598-018-26510-5

Here is another attempt to categorize smells. I add this because they bring up a good point:

1. odor space is hierarchical, and

2. We first must separate smells into good/bad, and only then separate further into the dimensions of odor space.

I might alter that slightly and suggest that the first separation can be either good/bad or familiar/unfamiliar; the latter might be even more important in categorizing smells.

(Note that another important part of this study in particular is about how trigeminal sensations influence the way we organize smells in the brain, and that this is one of the many things that make it such a messy task.)