Belgian scientists train AI to predict the taste and quality of beer

01 Apr 2024 --- Belgian scientists have unveiled an AI method for beer analysis with models capable of predicting consumer ratings of beers and identifying the aroma compounds that could enhance their flavors.

Published in Nature Communications, the study looks into the possibility of understanding and predicting the flavor of beer — a complex challenge owing to the intricate interplay of chemical compounds and subjective taste preferences.

Traditional methods, reliant on descriptive guides and personal tastes, fall short of providing a scientific basis for flavor comparison. Seeking to transcend these limitations, scientists have used chemical analysis techniques to study the flavor profiles of beers with unprecedented precision.

Kevin Verstrepen, professor at KU Leuven and director of the VIB-KU Leuven Center for Microbiology and the Leuven Institute for Beer Research, tells Food Ingredients First: “Our model went a step beyond because it can better predict the effect of the complex interactions between individual compounds.”

“This shows, for example, that some of the aroma compounds that are generally considered as negative off-flavors, like ethyl acetate, dimethyl sulfide, diacetyl and certain aldehydes, can actually have a positive effect on beer aroma when they are present in lower quantities and co-occur with other compounds like the fruity acetate esters.”

Machine learning methods
Over five years, scientists dissected the chemical makeup of 250 distinct Belgian beers. They measured over 200 chemical properties of each beer and combined these findings with sensory analysis conducted by a panel of trained experts.

The key innovation came from the application of machine learning to this vast array of data. The research team mapped data from over 180,000 consumer reviews and trained ten different machine learning models, with Gradient Boosting emerging as the most effective. This model outperformed traditional statistical predictions, accurately linking chemical profiles to complex food features and consumer appreciation.

The AI’s analysis highlighted specific compounds previously underestimated in their impact on beer flavor and appreciation. By adding these compounds to commercial beers, the researchers achieved variants with improved consumer ratings.

Verstrepen suggests that further refinements can be made to better model individual taste preferences as well as the influence of the environment, mood and age of consumers.

Healthy beer
Consumers are increasingly adopting low and no-alcohol options, prompting beverage innovators to develop adult-tasting alcohol alternatives that prioritize both health and flavor, driven by trends favoring moderate alcohol consumption and growing social acceptance of these beverages.

“Our biggest goal now is to make better alcohol-free beer. Using our model, we have already succeeded in creating a cocktail of natural aroma compounds that mimic the taste and smell of alcohol without the risk of a hangover,” says Verstrepen.

Last winter, Belgium-based beer company Thrive introduced a non-alcoholic beer fortified with ten essential vitamins, including 50% of the daily recommended intake of vitamin D3 and all B vitamins per can. This beverage offers nutritional benefits while boosting the immune system and reducing fatigue, with fewer calories compared to traditional beers.

“We see a lot of room for progression with those beer products, mimicking the alcohol aroma. But the general methods are certainly also applicable to any other food, although this would require to again analyze and taste a lot of samples,” concludes Verstrepen. 

By Sichong Wang