America is a country that loves dieting. Nearly half of Americans have tried to lose weight in the past year, a CDC study found, while other data suggests as many as 80 percent of Americans have tried dieting at some point in their lives. About half of U.S. adults have a diet-related chronic disease, such as cardiovascular disease, high blood pressure, or type 2 diabetes, so what’s at stake is far more than vanity — it’s individual and public health.
Yet who among American dieters hasn’t thrown up their hands at the way rich and fatty foods seem to make no impact on French waistlines, or wondered how other traditional diets seem to be so uniquely blessed with mysterious health benefits?
Here to bridge these enduring gaps in our food knowledge is UC, and specifically, the USDA-NIFA AI Institute for Next Generation Food Systems (AIFS) at UC Davis. One of seven AI institutes established by the National Science Foundation in 2020 (there are now 29), researchers at AIFS have been focusing on applying the latest computing tools to all aspects of the food system, from crop breeding to food safety to student outreach about AI-integrated agricultural careers. Among its most exciting projects is FoodAtlas, an effort that is using artificial intelligence to bring together scientifically validated information about food from papers published over the course of 80 years. It’s an effort leveraging the resources of UC Berkeley, UC Davis and UC Agriculture and Natural Resources that could revolutionize food science — and finally help us understand what makes our food, and our bodies, tick.
Exploring ‘nutritional dark matter’
One of the most surprising aspects of food science is that, despite the size of the nutrition industry, valued at more than $100 billion in 2024, we don’t actually know that much about food. The average consumer is probably familiar with calories, and some vitamins and minerals — a tiny sliver of what makes up our food. Even the U.S. Department of Agriculture only tracks about 150 chemical compounds in food, whereas researchers estimate that the average food item contains anywhere from 20,000 to 50,000 compounds. Scientists have taken to calling the enormous amount we don’t know about food “nutritional dark matter.” Unlike the dark matter in space, this dark matter has immediate impacts on our health.
Initiatives like the Periodic Table of Food Initiative (PTFI) in the United States (another effort led by UC Davis; see sidebar) and FooDB in Canada are supporting researchers as they expand our understanding of the chemical compounds in our food, the origins of those compounds, and their hidden health effects. FoodAtlas, meanwhile, is drawing a knowledge graph of this information from scientific literature — connecting the dots of food, chemistry and health with evidence like never before. Putting the full power of science and artificial intelligence behind the idea that food is medicine, these combined efforts could deliver on the promise of diets that dramatically improve health while responding to our own individual environments and biomarkers as those too become more widely understood.
Cracking the Mediterranean diet
An example of a diet that science has helped us understand is the Mediterranean diet. Its effects have been widely documented — its healthy fats and whole, plant-based foods lower our risks for common chronic conditions, particularly heart disease. But what is the mechanism driving these effects? It may well be DMB, or 3,3-dimethyl-1-butanol, found in olive oil, red wine, grapes and raisins, the Cleveland Clinic found, raising the notion it could even be used as an anti-heart disease pharmaceutical. That it is naturally occurring in food goes straight to the heart of the idea that food is medicine — and that if our knowledge can catch up with what we eat, we could offer health interventions that are accessible and delicious.
One easy win FoodAtlas has already found is that we have access to a lot more knowledge about foods than we had previously collected. Valuable nutritional information can make it into journals but get stuck there, never making it out to the public.
That’s where the artificial intelligence aspect of FoodAtlas comes in. FoodAtlas synthesizes published information from reputable journals and links the studies, the compounds, and the foods containing them neatly together. Nearly half of the food information FoodAtlas was able to gather from scientific journals hadn’t made it into a database anywhere. Networking all available evidence and studies about the foods we eat can be a force multiplier for the research presently being done to expand our understanding of the compounds in food. And it brings us increasingly closer to a world where health and nutrition fully meet — where a doctor can tell you what foods to put in your diet that respond to individual biomarkers or health conditions.
"Knowing what chemicals are in our food and what they do to our bodies is essential for public health," says Ilias Tagkopoulos, director of the USDA-NIFA AI Institute for Next Generation Food Systems (AIFS). "FoodAtlas has been created with this mission in mind: an AI system that reviews the published literature and databases to map foods, their molecular composition, their health effects, and other parameters that are important for decision support in creating healthier, more sustainable food."
Swap it Smart
While it’s good to know scientists are making up ground in understanding the compounds in our food, absorbing all that information as a consumer may seem daunting. For all its 90s simplicity, a nutrition label is often too much to take in, let alone figure out how the compounds one label describes interact with those on the label of another food in your grocery cart. Every trip to the grocery store is, fundamentally, a scientific experiment you’re conducting on your own body while trying to finish other errands and make ends meet. Fortunately, UC Davis is developing an app to bring the power of the FoodAtlas and the Periodic Table of Foods Initiative (PTFI) to consumers.
Called “Swap it Smart,” the app will bring AI-powered food composition science into meal formulation. The app categorizes meals across food quality, nutrition, bioactivity, environmental impact, affordability, and flavor while generating recipes aligned with the user’s health and sustainability goals. The ingredient substitutions (“swaps”) suggested will rely on advanced understanding of how food items impact health and the planet. UC Davis already has plans to collaborate with renowned California chefs who will be using the app to make recipes that meet their kitchens’ own nutritional and sustainability goals — expert validation of all that the app, and FoodAtlas, can do.
The app categorizes meals across food quality, nutrition, bioactivity, environmental impact, affordability, and flavor while generating recipes aligned with the user's health and sustainability goals. The ingredient substitutions (“swaps”) suggested will rely on advanced understanding of how food items impact health and the planet.
“Swap it Smart is building on the strengths of FoodAtlas and PTFI to create a platform for meal development,” Tagkopoulos says. “It is designed to help nutritionists in schools and chefs in restaurants make delicious meals and dishes that are good for our health and the environment.”
Not every compound scientists discover in our foods will prove consequential — many are likely to simply pass through our bodies with no effect. Some are likely to have beneficial health effects for only some populations — validating the experience many dieters have of success with some combinations of foods but not others. Scratching our heads over what’s included and excluded from a suggested diet may soon be a thing of the past — enjoying tasty and delicious food, decidedly not.
