From odour molecule to affect: auditing the four-tier evidence chain across 52 compounds

Objective. To test how far current knowledge allows us to go from a specific odour molecule to a predictable emotional effect in humans, by building a four-tier profile for each compound: molecule → human receptor → neural code → measured emotional effect.

Methods. Four parallel collection streams, sharing a single 18-field schema and four strict rules: only human data accepted for the neural and emotional tiers; concentration and mode of odour presentation must be recorded; pure molecules must be distinguished from essential-oil mixtures and from oral preparations; conflicts of interest must be recorded. The results were merged into a database of 52 single molecules, cross-checked across the four streams.

Main results. The four-tier causal chain is complete for only 5 of 52 molecules (carvone, menthol, linalool, trimethylamine, vanillin), and even for these five each tier rests on one to three small studies. At the receptor tier, about 88% of human odorant receptors still have no identified ligand, and about 17% of deorphanisation results fail to replicate when the assay system is changed. At the neural tier, chemical structural features do not predict activity patterns in the human olfactory cortex, whereas perceptual features do; CO₂ — an odourless substance — activates the piriform cortex with every measurement method; and the same chemical given two different labels produces two different brain responses. At the emotional tier, no molecule-level meta-analysis exists, and not a single inhaled single-molecule study reports confidence intervals. When trials with fewer than 30 people per arm are excluded, the pooled effect of aromatherapy on the STAI-State anxiety scale falls from −6.63 (95% CI −8.17; −5.08) to −0.84.

Usable quantitative regularities. People can resolve only about three components in a mixture, and training does not raise this limit; from about 20–30 equal-intensity components spanning the odour space upwards, different formulas converge on the same smell. The valence of a mixture is an intensity-weighted average (R² = 0.87–0.95), not a pleasantness-weighted one. The concentration–emotion relationship operates on the number of steps above each individual’s threshold (|ρ| = 0.79–0.95), not on physical concentration. Olfactory adaptation brings perceived intensity down to an asymptote of 25–40% after about four minutes and then stops, rather than falling to zero.

Decomposition of valence variance. Individual differences account for about 54% of the variance, molecular identity 14–41% depending on the odour set, familiarity 37%, and culture only 6–7%. Pleasantness rankings transfer well between markets (r = 0.82), but absolute score levels do not.

Conclusion. In the current state of knowledge, it is not possible to go reliably from a chemical formula to a predetermined emotional state. What can be controlled lies not in the molecule but in three variables with large, measurable effects: learned association, context and label, and intensity relative to the smeller’s threshold. The report turns the measured regularities into design principles, and lists 20 popular claims that the data refute or that are misattributed to their source.