Flavor perception is incredibly complex

The origin of our sense of flavor predates the rise of animals. Even single-celled organisms navigate their world by sensing chemicals in their environment. Our sense of taste evolved for survival, by detecting toxins (bitter), spoilage (sour), energy (sweet), and nutrients (umami and salty). Our sense of smell amplifies our taste with a vast bouquet of odors at an incredible level of sensitivity, providing the necessary context to our tastes.

Flavor perception, or how our brain interprets the signals from our nose and mouth, is incredibly complex. Today, over 300 flavor receptors have been mapped. Compare this to our sense of sight, operating on just 3 color sensors, and you realize that every single person would fail a “color-blindness” test on flavor. These genetic loss-of-function mutations, called “anosmias”, or changes in receptor sensitivity are fundamental to our differences in flavor perception: they are the reason why we have different favorite foods.

But our receptors are just part of the picture. Flavor preference, how much we like and dislike foods, is also strongly influenced by exposure: our brains learn to love or hate flavors based on experience, and even become increasingly sensitive with frequent re-tasting. This is influenced by our culture, dictating the foods we are commonly exposed to and the settings in which we consume them. If the setting is not quite right, our mental state comes into play. We only like flavors if we are in the mood for them. We enjoy flavors more when we are happy and relaxed than when we feel anxious or stressed.

A final layer of complexity occurs when we perceive flavor. Our brains do not directly translate messages coming from our receptors. Instead, they look at the whole array of messages from our senses and create a unified response of flavor perception. This allows flavor-active chemicals to interact and compete for our attention. Consequently, flavor molecules amplify or mask each other through synergistic or antagonistic effects. This is fundamental to cooking, where we combine different raw ingredients to create flavorful dishes. We create mixtures of flavor compounds that work in harmony.

Using humans as analytical instruments

This multi-layer complexity of flavor makes food science very challenging. So how can food manufacturers ensure their product is going to be successful? Their best option is to use humans, trained expert tasters, to evaluate new products and perform quality control. This works because (1) there are general trends in flavor preference and (2) the effect of individuals gets averaged out by multiple tasters. The former arises from evolution, as natural selection favors humans who innately like flavors that are tightly linked to nutritious meals and dislike flavors that are linked to harmful substances.

Using humans works well, but has its disadvantages. First, there is the issue of averaging out: foods that are good on average are not necessarily the best food for certain individuals. Manufacturers must decide whether they want to make an averagely good product or a great product marketed toward specific consumers, and their tasting panel should reflect this target audience. Then there is the issue of exposure, a trained expert’s sense of flavor adapts as they repeatedly assess similar products. Consequently, during product development, the panel can acquire a taste for the developed product that new consumers do not share.

Lastly, panels must be sufficiently large and diverse to cover common genetic diversity, or you risk missing out on a large part of the public. If you market your product towards international or global markets, your tasting panel should reflect this cultural diversity. The requirement for sufficiently large panels leads to the biggest concern: costs. Running a tasting panel is expensive, time-consuming, and low throughput.

This raises the question of alternatives. If flavor receptors are essentially analytical instruments, can modern chromatography do better? Predicting flavor from chemical composition has been a holy grail in sensory science for decades, and steady progress has been made over the years. Nowadays, most (if not all) food manufacturers measure process parameters that steer the production process, alleviating some pressure on sensory panels. Still, accurately predicting the flavor of complex foods remains elusive.

The next generation of food research

That is where our work comes in. We firmly believe that modern advancements in artificial intelligence (AI), combined with ever-improving analytical capabilities, will finally bridge the gap from food chemistry to flavor. In our recent paper “Predicting and improving complex beer flavor with machine learning”, we highlight the potential for machine learning and AI in food science, showing a marked improvement over conventional statistical models. Our work is a convincing proof-of-concept but only scratches the surface of the potential for AI in food research.

One of the biggest hurdles in the years to come is gathering large datasets that can properly train computer models. On the chemical side, this requires innovations towards fast, high-throughput, untargeted, and highly sensitive analytical instruments. Recent innovations in NMR, direct-MS/MS, and ion-trap-MS are making great progress on that front. On the sensory side, consumers could be key for amassing vast amounts of data. By providing their flavor preferences in online food reviews, consumers provide sensory scientists access to invaluable datasets. This is already happening in the wine and beer industry through review platforms like Vivino and RateBeer. The ease of access and sheer size of these databases provide amazing new opportunities for food research. The final challenge will be translating these new insights into food products. Modifying the chemical composition of beverages is straightforward - complex food matrices, not so much. We are excited about the challenges that lie ahead!

Flavor perception is more complex than you might expect, but things are moving in the realm of sensory science. Like many fields, food research is entering the world of big data and artificial intelligence. In part, this shift is made possible by the voluntary contribution of consumers, without whom our research would not have been possible. So next time you enjoy a meal, leave a review, you are actively contributing to the taste of the future.

The authors thank Julian Ernesto Prieto Vivas, Sasha Yogiswara, and Jonas D'hooge for feedback, suggestions, and proofreading.