Decoding hidden wine patterns: how mineral profiles and machine learning reveal identity clues for recognising grape varieties beyond sensory perception

Abstract

Previous sensory studies have consistently shown that both professionals and consumers encounter difficulties distinguishing wines consistently according to expected categories, such as grape variety or appellation. This raises the question of whether specific sensory characteristics can be used to discriminate between close grape varieties and appellations [MD1.1]. In this exploratory study, we analyzed the Mineral Wine Profiles [1] (40 elements through inductively coupled plasma mass spectrometry (ICP-MS)) of twelve red wines (six Gamay and six Pinot noir from the 2011 vintage) that had previously been evaluated in a sensory categorization task. We then submitted these mineral profiles to a machine learning model trained exclusively on several thousand wines from recent vintages. The model had learned to recognize varietal signatures from multi-element combinations and to capture complex relationships among elements rather than relying on a single marker. This approach makes it possible to objectively assess whether elemental composition alone contains sufficient information to classify wines according to grape variety. While wine professionals showed inconsistent varietal grouping between the 12 wines, the model achieved a perfect grape varietal prediction, showing a clear and robust separation between Gamay and Pinot noir, relying on multi‑element combinations (especially Sr, Ba, S, Cu, Mn) rather than a single discriminant element. Together, these findings suggest the existence of robust, multi-element signatures that underline wine identity. They also demonstrate the potential of combining mineral profiling with machine learning to discriminate grape varieties, thus opening new perspectives on the potential relationships between chemical markers and perceptual expertise in wine evaluation.

References

Sarlo, L., Duroux, C., Clément, Y., Lanteri, P., Rossetti, F., David, O., Tillement, A., Gillet, P., Hagège, A., David, L., Dumoulin, M., Marchal, R., Tillement, T., Lux, F., & Tillement, O. (2024). Enhancing wine authentication: Leveraging 12,000+ international mineral wine profiles and artificial intelligence for accurate origin and variety prediction. OENO One, 58(4), 4. https://doi.org/10.20870/oeno-one.2024.58.4.8107