From chemistry to perception.
Wine holds thousands of interacting compounds, but measurement alone doesn't explain the experience of tasting it. Our research explores how structured wine data, a sensory model, and individual preference can be connected — without reducing taste to a universal score.
Research preview. Figures on this page are illustrative unless a source is cited.
Can a wine's measurable chemistry be connected to how a specific person is likely to perceive and enjoy it — honestly, with visible uncertainty, and without pretending chemistry determines quality?
That question sits between two facts. First: a wine's aroma and structure are the sum of real, measurable molecules. Second: what any one person likes is shaped by biology, context, and expectation. Oenra models the mapping between them — and is careful about what that mapping can and cannot claim.
What we read.
Different instruments see different parts of a wine. Where a producer or lab provides data, Oenra normalizes it into one structured profile.
Where a producer or lab provides it.
GC-MS resolves volatile aroma compounds; LC-MS the non-volatile phenolics, anthocyanins, and tannins; FT-IR estimates bulk composition through calibrated models. Wine contains more than a thousand volatile compounds, but only a few dozen are typically odor-active.
A wine becomes a set of numbers.
Chromatography spreads a wine's volatile molecules across time; each peak is a family of compounds. Together with non-volatile and bulk measurements, they form a structured, machine-readable representation of the wine.
A stylized chromatogram: intensity over retention time, with peaks labeled by compound family.
From inputs to an explained match.
Inputs
Analytical chemistry, producer specifications, vintage, region, varietal, and — where available — sensory-panel and expert descriptions.
Wine representation
Inputs are normalized and combined into a structured, machine-readable representation of the wine.
Sensory representation
That representation is expressed across eight sensory dimensions with per-dimension confidence.
Preference representation
A person's stated preferences and feedback become a taste model on the same dimensions.
Matching
The two profiles are compared to produce an explained match — not a claim about universal quality.
Evaluation
Predictions are checked against real preference feedback and blind comparisons as the dataset grows.
A dumbbell plot comparing a wine's eight-dimension sensory profile against a personal preference profile on a shared scale.
- This wine
- Your palate
Two profiles, one scale.
The wine's representation is expressed across eight sensory dimensions, each with a confidence. A person's stated preferences and feedback become a taste model on the same dimensions. Matching compares the two — a chosen dot plot rather than a single distance, so the reasoning stays visible.
What the model can't do.
This section matters as much as the rest. A responsible system is honest about its edges.
Chemistry predicts some things well, others poorly
In peer-reviewed studies, instrumental chemistry plus chemometrics predicts a few composition-linked sensory attributes moderately well and many aroma attributes weakly. There is no basis for a single “we predict taste with X% accuracy” number, and we don't claim one.
Measured quality is not personal enjoyment
Flavor is constructed by the brain from taste, smell, vision, expectation, price, and context. In blind tastings, non-experts do not enjoy more expensive wine more (Goldstein et al., 2008); stated price alone can raise reported pleasantness (Plassmann et al., 2008). Chemistry under-determines who will like a wine.
Context changes the wine
Serving temperature measurably shifts perceived aroma and taste; even color and labeling change how a wine is described (Morrot et al., 2001). The same bottle is a different experience across settings.
People differ, biologically
Genetic variation — for example in the TAS2R38 bitter-receptor gene — changes how intensely a person tastes certain compounds. Effects in wine are real but modest, and don't reduce to a single number.
Confidence depends on the data
A wine with rich analytical chemistry gets a narrower estimate; one identified only from its label gets a wider one. We show the confidence rather than implying certainty we don't have.
What's real, and what's ahead.
Oenra is a research preview. Here's an honest read on where each capability stands.
- GC-MS
- Gas chromatography–mass spectrometry. Separates and identifies a wine's volatile (aroma) compounds.
- LC-MS
- Liquid chromatography–mass spectrometry. Targets non-volatile molecules — phenolics, anthocyanins, tannins — that shape color, taste, and mouthfeel.
- FT-IR
- Fourier-transform infrared spectroscopy. Rapidly estimates bulk composition (alcohol, sugars, acidity) via calibrated models — a prediction, not a direct read of trace aromas.
- Volatile compound
- A molecule that evaporates readily and can reach the nose, producing aroma. Wine holds more than a thousand; only a few dozen are typically odor-active.
- Phenolic
- A class of plant compounds (including anthocyanins and tannins) responsible for a wine's color, astringency, and structure.
- Chemometrics
- The use of statistics and multivariate models to extract information from chemical measurements.
- Sensory analysis
- The structured evaluation of a product's attributes by human senses, using a standardized vocabulary.
- Confidence interval
- A range that expresses how certain an estimate is. Wider means less certain — usually because there is less data.
- Preference model
- A model that relates what a person likes to the sensory characteristics of products, to identify the drivers of their preference.
- 01Kopsacheilis et al. (2024). Crowdsourcing the assessment of wine quality. Journal of Wine Economics 19(3).
- 02Hodgson, R.T. (2008). An examination of judge reliability at a major U.S. wine competition. Journal of Wine Economics 3(2).
- 03Goldstein et al. (2008). Do more expensive wines taste better? Journal of Wine Economics 3(1).
- 04Plassmann et al. (2008). Marketing actions can modulate neural representations of experienced pleasantness. PNAS 105(3).
- 05Morrot, Brochet & Dubourdieu (2001). The color of odors. Brain and Language 79(2).
- 06Spence, C. (2020). Wine psychology: basic & applied. Cognitive Research: Principles and Implications 5:24.
- 07Kim et al. (2003). Positional cloning of the human quantitative trait locus underlying PTC taste sensitivity. Science 299(5610).
These are external scientific sources that inform our thinking. They are not claims that Oenra has independently reproduced their results. See the Journal for our own writing.
Understand the bottle. Understand the person.
Oenra is a research preview. Join to help shape it, or talk to us about a pilot.