Guided by a machine learning model that predicts what makes Belgian beer taste more appealing, researchers tinkered with the composition of golden nectar and greatly impressed participants in a blind tasting. This method could lead to new recipes and improved tastes for a variety of foods and beverages. Sante!
Controversial opinion: I’ve never been a big beer drinker. I don’t like the taste. But as an Australian, I’m sure it’s hardwired into all of us from birth that beer is considered by many to be the golden nectar.
Understanding and predicting how we perceive and value what we consume, such as beer, is a major challenge for the food and beverage industry. Indeed, it makes commercial sense for the industry to induce freaks like me to drink (or eat) its products. Researchers at the University of Leuven in Belgium have developed a machine learning model to develop beer flavors that are more appealing to consumers and help manufacturers meet specific consumer needs.
“Predicting flavor and consumer evaluation from chemical composition is one of the ultimate goals of sensory science,” the researchers said. “A reliable, systematic and unbiased method of linking chemical profiles to flavor and food evaluation would be an important asset to the food and beverage industry.”
First, to generate a comprehensive dataset on beer flavor, the researchers selected 250 commercially available Belgian beers from 22 different beer styles. The majority of the dataset consists of blondes (12.4%) and tripels (11.2%), reflecting the presence of blondes in the Belgian beer scene and the diversity of beers in these styles. The researchers then measured 226 different chemical properties of each beer, including brewing parameters such as alcohol content, pH, sugar concentration, and more than 200 flavor compounds.
A trained tasting panel evaluated and scored each of the 250 beers on 50 sensory attributes, including various hops, malts, yeast flavors, off-flavours, and spices. To expand on the tasting panel data, researchers collected 180,000 reviews of 250 selected beers from consumer review platform RateBeer. This yielded numerical scores for appearance, aroma, taste, taste, overall quality, and overall average score.
The researchers used a combination of chemical analysis, tasting panel ratings, and public reviews to train a machine learning model. They then leveraged the model to infer the important contributors to sensory perception and consumer evaluation, recognizing that products with poor consumer ratings are not commercially successful.
Ethyl acetate was identified as the most predictive parameter for beer rating. It usually has a fruity, solvent, and alcoholic flavor. Ethanol, the most abundant beer compound after water, was the second most important parameter. Ethanol not only directly contributes to beer flavor and mouthfeel, but also greatly influences the physical properties of the beverage and determines how volatile compounds contribute to aroma. Lactic acid, which contributes to sour beer’s acidity, was also highly praised. Interestingly, some of the most important predictive parameters are not the well-established beer flavor, which is generally associated with negative beer quality. For example, ethylphenyl acetate, which is commonly associated with beer aging, was found to be an important factor contributing to beer evaluation.
Finally, the researchers tested whether the predictive model provided insight into beer ratings. They specifically selected overall evaluation as the factor to investigate due to its complexity and commercial relevance. Because adding a single compound can make a noticeable difference that unbalances the beer’s flavor profile, the researchers evaluated the impact of changing the combination of compounds. We also chose blonde beer as the starting material for our experiments because it was strongly represented in our dataset.
Adjusting for concentrations of ethyl acetate, ethanol, lactic acid, and ethylphenyl acetate, the most important predictors of overall rating, significantly increased overall rating compared to control among a panel of trained tasters. improved. Panelists noted an increase in flavor intensity, sweetness, alcohol, and body fullness. To remove the contribution of ethanol to the results, we performed her second experiment without the addition of ethanol. Similar results were obtained, including higher overall ratings. Further experiments tested whether the model’s predictions could improve ratings of non-alcoholic beer. Again, a mixture of predicted compounds excluding ethanol was added, resulting in a significant increase in taste, body, flavor, and sweetness.
“Our study confirms that the concentration of flavor compounds does not necessarily correlate with perception, suggesting complex interactions that are often overlooked by traditional statistics and simple models. ” said the researchers. “The predictions of our final model, trained on review data, were tested by a small group of trained testers, as demonstrated by our ability to validate specific compounds that drive beer flavor and rating. This also applies to blind tastings conducted by
Researchers are mindful of the social burden caused by alcohol abuse and addiction and warn against using their methods to exacerbate this.
“We encourage the use of our results for the production of healthier and tastier products, such as new and improved beverages with lower alcohol content,” they said. “Furthermore, we strongly discourage the use of these techniques to improve the recognition or addictive properties of hazardous substances.”
They hope that future research will be able to expand the scope of the study to include more diverse markets and beer styles, and identify more rating factors.
“Soon, these tools may offer solutions in quality control and recipe development, as well as new approaches to sensory science and flavor research,” they say.
The study was published in the journal nature communications.
Source: University of Leuven, via Scimex
