In short
Structure-odor relationships ask how a molecule's shape and chemistry relate to its smell. There are useful partial rules, but no general theory that predicts a new molecule's odor reliably. Tiny changes can transform a smell, while unrelated structures can smell alike. Recent studies predict some perceptual attributes with useful accuracy, but they have not solved the problem.
What you will learn
- Explain why structure-odor prediction is hard
- Give examples of similar molecules that smell different, and the reverse
- Describe what recent prediction studies do and do not show
Before this lesson: Lactones, Ketones and Macrocycles
What makes it hard
- Many receptors. Each odorant can activate several receptor types, and each receptor can respond to several odorants. Identity is a pattern.
- Perception is not chemistry. Between receptor and description come neural processing, memory, language and culture.
- Individuals differ, so there is not one true answer for every molecule.
- Impurities matter. A smell attributed to a molecule might come from a trace impurity in the sample. Serious studies try to rule this out, and that is difficult.
- Concentration changes the smell. Indole is the familiar example.
Three kinds of case
| Case | Example |
|---|---|
| Similar structure, different odor | The carvone enantiomers: l-carvone is described as spearmint, d-carvone as caraway |
| Small change, big effect on strength | Ionones and damascones, as you saw in the previous lesson |
| Unrelated structures, similar odor | Macrocyclic, polycyclic and nitro musks |
What partial rules exist
Within narrow families, chemists have long found patterns. In a series of aldehydes, for instance, the smell shifts in a broadly predictable way with chain length. In particular musk or sandalwood families, researchers have identified features that seem important for the odor. These local rules are useful in designing new materials, even if they do not extend to all odorants. The molecular perfumery track looks at such families in more depth.
What recent prediction work shows
Two studies are worth knowing, with their limits.
Keller et al. (2017) ran a public challenge. Teams were given perceptual ratings from 49 people on 476 molecules and tried to predict ratings from chemical features. Models could predict several attributes, notably intensity and pleasantness, reasonably well on average. Yet this is prediction in a specific dataset, using specific descriptors. It does not give you a formula for designing a molecule, and individual-level predictions were much weaker than the average.
Snitz et al. (2013) built a model predicting how similar two odor mixtures would seem, based on structure, with a good correlation in their tests. Again, this describes particular mixtures and a similarity judgment.
A sensible reading: prediction from structure has improved a good deal, particularly at the level of averages. But a perfumer still has to smell the material, and a supplier's claim that a molecule was designed by computer tells you little about how it will work in your formula.
Practical consequences
- Do not choose materials from a drawing. Choose them from your own strips.
- Do not assume that a chemical relative will dose alike.
- Do not trust "this molecule smells of X because of Y" unless you can see the evidence.
- Use local rules as a way to plan what to smell next.
Practical exercise: Test a structure-odor rule on your own materials
Use a family you can obtain, such as esters of one alcohol, lactones or fatty aldehydes.
- Pick three members of the family, and look only at the structures.
- State a prediction: which will be strongest, and how the character should shift.
- Smell them at matching dilutions, ideally blind, then mark your prediction right or wrong.
- Read what the encyclopedia says, and note any disagreement with your own reading.
You have done this well if your prediction was written first, and your note says what you can and cannot generalize from three molecules.
Common mistakes
- Treating a pattern as a law. Patterns hold within families and fail between them.
- Believing a single famous example explains everything. Carvones are striking partly because many other enantiomer pairs are much harder to tell apart.
- Quoting a machine-learning result as if it predicted your own nose. Averages are not individuals.
Key terms
- Structure-odor relationship
- The study of how a molecule's structure relates to how it smells. Often shortened to SOR.
- Odorant
- A molecule that can be smelled. Molecules must be volatile enough to reach the nose in order to be odorants.
Knowledge check
Written and reviewed by the RUŌOD Lab team. This is general education, not legal, regulatory, medical or safety advice. Example dilutions and exercises are for learning, are not validated products, and have not been safety-assessed. Completing a lesson confers no qualification. How we check these lessons.