Recap: Tasty, Nutritious and Filled with Complexity: Giulia Menichetti on how AI can Decode our Food

By: David Bolton

The idea that “you are what you eat” has been a popular saying for more than a century, but the essence of nutrition is about a lot more than mere lifestyle choices. The nutrients you need to function impact every part of your body, from skin and hair to muscle tone, bone conditions, immune function, and digestive health. That means your diet can be either a positive, enriching experience or the catalyst for myriad health problems.  

Giulia Menichetti, faculty member at Harvard Medical School and a faculty member in the Network Science Institute at Northeastern University, believes our nutritional health is largely shaped by mysterious forces: the unknown (or disguised) chemical composition of food. Her work as an associate research scientist on the Center for Complex Network Research’s Foodome Project is a clear sign of her commitment to nutritional research. To date, she has applied systemic and computational approaches to finding the connection between individual lifestyle choices and rates of coronary heart disease.

Menichetti admits she is not a computational scientist by trade, but she is interested in how food affects our wellness and how chemical interactions affect our bodies. The vast majority of her published work dissects the complex nature of food systems—the interconnected networks of trade and agriculture that impact human health and development. Her understanding of the unique nature of individual nutrition systems is one of the reasons why the Rockefeller Foundation and Acumen Academy selected her to be an inaugural cohort for the Food Systems Fellowship.

Decoding Complex Systems

A guest lecturer in the Fall Seminar Series hosted by the Institute for Experiential AI (EAI), Menichetti argued that the mechanistic relationship between food chemistry and human health is best explained through data. Quantifying the many different chemical actors present in food allows researchers to identify which chemical-heavy foods or packaged products people should stay away from. All that data can even point to new drug possibilities.

“There are extensive records that describe [food additives’] presence in food, with quantifiable amounts and interaction with known human protein targets,” Menichetti said, referring to the specific proteins known to respond to pharmaceutical intervention. “Unfortunately, we realized that this was not a realistic perspective to work from. To get a full picture, we needed to investigate the dark matter of nutrition.”

By using a term more commonly associated with cosmology, Menichetti alluded to the unmapped and often mysterious chemistry of food, as well as the thousands of nutrients that the United States Department of Agriculture (USDA) has yet to identify. The UDSA currently tracks and catalogs around 150 key nutritional components, whose varying caloric, sugar, fat, and vitamin contents reveal noticeable effects on human health. By contrast, the Foodome database has detected 88,747 compounds and inferred that a further 46,484 exist across the food universe.

But to get a true perspective of how complex food systems really are from a nutritional perspective, you need to bring network science into the equation. Menichetti’s mission to track the complexity of food systems with AI then becomes a study in how to map the interactions between food, biology, nutrition, and data to improve health outcomes and lifestyle choices.

The caveat, however, is that most actionable information is not readily available. Researchers have to cast a wide net to get the data they need, with public, private, and academic resources all providing their own chokepoints. To work around these information silos, you need a state-of-the-art tool that can collect and analyze disparate knowledge bases. That tool, she says, is artificial intelligence.

“When we were writing this perspective using open-source databases, we found around 26,000 chemical compounds and small molecules,” Menichetti said “These were sparsely distributed across different foods. After significant efforts in data integration—which included combining the ambiguities and annotations from existing scientific literature—and the aggregated databases associated with genomic and pathway interactions, we recently reached 135,000 chemical compounds. And we keep growing.”

Data Driven, Questions Answered

In Menichetti’s opinion, questions about the overall nutritional value of a particular food can best be answered through complex system modeling and by mapping the key chemical compounds, including their interactions within the body. Menichetti and her team have leveraged AI and deep learning to measure nutrient performance and reduce biases associated with disparate data. These results allowed her to build a comprehensive database from what she calls the “food data jungle,” an ecosystem of seemingly unaligned public health data that obscures what should be a straightforward question: How much actual nutrient value exists in food?

Consider, for example, so-called “ultra-processed food,” a category that includes food preservatives, dyes, and flavor enhancers that can make food (in theory) tastier and more convenient but rarely add any nutritional value. This differs from what we know as packaged or processed foods, in that the additives–salt, oil, sugar, for example–are intended to enhance the raw materials themselves. While ultra-processed food may promote convenience over fresh or packaged alternatives, they have been linked to increases in global rates of obesity and food-related heart disease.

With so many compounds to identify and evaluate, AI quickly becomes essential. Deep learning algorithms in particular are a tremendous research asset, especially when it comes to mapping protein and ligand binds that have not been seen before.

“There are definitely traditional methods, such as molecular dynamics or docket simulation that simulate this type of interaction,” Menichetti said. “Unfortunately, these algorithms have very large computational complexity, as they require exploring all the possible locations on the molecules. This is the reason why machine learning and AI provide us with an opportunity for more precise analysis.”

Understanding Nutritional Value = Better Health Outcomes

There is an argument to be made that the abundance of processed and ultra-processed foods has a radical impact on both physical and mental health. In recent years there has been an increased focus in the healthcare sector on food consumption habits, much of which has been directed at the real nutritional value of packaged and ultra-processed products. Food manufacturers have been labeling their products with Nutrition Facts since 1990, but the accuracy of the information can depend on the in-house testing that these brands are required to perform.

On a global basis, poor nutrition is a leading cause of death through heart disease, with the United States spending more than one trillion dollars on food-related health costs in 2021 alone. Throw in rising obesity rates—just under 40 percent of adults aged 18 and over were classed as obese in 2016, according to the WHO—plus a never-ending stream of slick marketing campaigns that put processed food front and center, and you can see why this type of data-driven research matters.

As long as dedicated network science specialists such as Giulia Menichetti are laser-focused on providing the transparency and information people need to make informed decisions about what they eat, then this is good news. And if you need more proof as to why we need serious conversations about nutrition and its relationship to complex systems, check out the online version of her database and search for cheesecake.

To find out more about Menichetti and her research, you can watch a replay of the talk here. And don’t forget to register for upcoming seminars!

You also can read more about why chemically-altered foods are not a good thing to eat or drink in this recent Northeastern article.