What if you could identify not only the aroma of whiskey but also its place of production through smell? A German research team has developed artificial intelligence (AI) that can evaluate the flavor profile of whiskey and accurately differentiate whether it is from the United States or Scotland with 90% accuracy.
The German research team led by the Fraunhofer Institute for Process Engineering and Packaging noted that they have developed two machine learning algorithms to discern the production location and flavor profile of whiskey, successfully distinguishing American whiskey from Scotch whiskey. The findings were published in the international journal Communications Chemistry on the 20th.
The flavor profile of whiskey is determined by a combination of dozens of different compounds. This is why it is difficult to predict whiskey's characteristics merely by analyzing odor molecules. While there are experts available to identify the main flavors of whiskey, the training requires significant time and expense, and differing opinions among experts present limitations.
The research team predicted the aromas of whiskeys using their self-developed molecular flavor prediction algorithm OWSum and artificial neural networks. Upon collecting and analyzing molecule data related to aroma, they found that American whiskeys showed high levels of menthol—a compound found in mint and other plants—and fragrant citronellol, which comes from flowers such as geranium and rose. In contrast, Scottish whiskeys primarily emitted methyl decanoate, a compound used as an artificial flavoring, and caprylic acid, which has an oily scent.
The algorithms referenced each compound to identify the characteristic notes of the whiskeys. OWSum quantified the importance of each compound, while the artificial neural network analyzed the structural patterns of the compounds. As a result, the primary notes of American whiskey were caramel, while those of Scottish whiskey were phenols, which give off an apple or musty smoke smell. The research team explained, 'Both algorithms were able to identify the strongest notes of specific whiskeys more accurately and consistently than human experts.'
Based on this, they conducted experiments to distinguish between seven types of American whiskey and nine types of Scottish whiskey. The OWSum algorithm developed by the team successfully classified whether a whiskey was American or Scottish with over 90% accuracy.
The research team stated, 'This algorithm can efficiently and quickly process the classification of the whiskey's production location and identification of major flavors,' adding, 'In the future, once we validate the versatility of the model using more diverse samples and data, it could contribute to analyzing flavors or automating quality control in the whiskey industry.' They also noted, 'It could help detect chemical changes that may occur in whiskey during production early on and determine irregularities, or assist in developing processes that maximize specific flavors.'
References
Communications Chemistry (2024), DOI: https://doi.org/10.1038/s42004-024-01373-2