Belgian Beer Study Incorporates Machine Learning for Recipe Development
Researchers from the Vlaams Instituut voor Biotechnologie (VIB) in Flanders recently delved into the world of Belgian beer with a new approach – machine learning. Led by doctoral student Michiel Schreurs, the team aimed to enhance the flavors of both alcoholic and non-alcoholic beers to cater to consumer preferences better.
Beer chemistry and taste are intricately connected, making it a challenge to predict how different chemical properties influence the overall flavor. Traditionally, this process often involves trial and error along with extensive consumer testing. However, the Belgian group decided to take a different route.
In their study, the researchers analyzed 200 chemical properties from 250 Belgian beers spanning 22 styles, such as Blond and Tripel beers. By leveraging the RateBeer online database and sensory descriptions, they linked chemical compositions to flavor profiles. Additionally, data from a trained tasting panel of 16 individuals further enriched their dataset.
Using this wealth of information, the team trained machine learning models to predict flavor preferences based on the chemical makeup of the beers. The results, published in Nature Communications, demonstrated that the machine learning model outperformed human testers in blind tastings in terms of overall appreciation.
Schreurs emphasized the complexity of beer flavors, noting that it would be impossible to predict taste based on a few compounds alone. By harnessing the power of computers, the researchers were able to gain valuable insights into the relationship between chemical properties and consumer appreciation.
The authors believe that this approach could revolutionize quality control and recipe development not only for beers but also for various food and beverage products. While their focus was on Belgian beers, they are optimistic about the broader implications of their findings.
One of the key goals moving forward is to enhance the quality of alcohol-free beer. Professor Kevin Verstrepen from KU Leuven highlighted the team’s success in creating natural aroma compounds that mimic the taste of alcohol without the side effects, thanks to the machine learning model.
As the researchers celebrated their achievement with a round of Belgian beers, their work paves the way for innovative advancements in the beverage industry. By combining traditional brewing techniques with cutting-edge technology, they are well-positioned to cater to evolving consumer demands and preferences.
