Skip to Content
Artificial intelligence

Flavor Networks Reveal Universal Principle Behind Successful Recipes

Hidden chains of flavors between ingredients explain what makes some dishes taste better than others, say food scientists.
In this graph representing the top 100 ingredients in the global flavor network, nodes are ingredients, edges represent shared flavors, and node colors represent clusters of linked ingredients.

Given the number of ingredients that humans eat, the total number of ways to combine them is on the order of 10 to the 15th power. And yet the actual number of recipes we eat is around one million—a small fraction of the total. That strongly suggests an organizing principle that, in recipe terms, sorts the wheat from the chaff.

So an ongoing challenge for food scientists is to discover laws that govern flavor combinations and use them to create new recipes yet to be experienced by human taste buds.

Today, Tiago Simas at Telefonica Research in Barcelona, Spain, and a few pals say they have discovered an important principle of flavor combination by studying foods of different cultures. This new insight could help create novel recipes.

The background to this group’s discovery is the hypothesis of food pairing developed by the chefs Francois Benzi and Heston Blumenthal. At first glance, foods such as chocolate and blue cheese can seem as different as it is possible for foods to be. And yet, these foods share 73 different flavor molecules.

That’s why at certain high-end restaurants, you’ll sometimes find blue cheese and chocolate in the same dishes. The thinking is that when ingredients contain the same flavor molecules, they can be successfully paired. The idea is that shared flavors help blend ingredients more effectively. Food pairing immediately suggests a novel way to create new recipes, which is why it rapidly gained influence among a certain breed of gastronomist.

Then in 2011, a curious piece of research revealed that food pairing was only part of the explanation behind successful recipes. In this work, a team at Harvard University in Cambridge, Massachusetts, analyzed the network of links between ingredients in recipes from all over the world. In this network, ingredients are nodes in a web, linked when they share flavor molecules.

This approach turned the food-pairing hypothesis on its head. When recipes from North America and Western Europe are analyzed in this way, the networks reveal that food pairing is an important factor. But when the team analyzed recipes from East Asia (Korea and Japan, for example), they found exactly the opposite. These cuisines seem to combine the very foods that do not share flavor ingredients. Clearly the food-pairing hypothesis is just part of a bigger picture and in need of a serious upgrade.

Enter Simas and his colleagues. These guys have looked a little harder into the web of flavors behind recipes and discovered a deeper principle at work. The basic idea is that when two ingredients do not share flavors, the team look for a third ingredient with flavors in common with each of the first pair. In this way, they were able to identify flavor chains and explore how recipes in different parts of the world use them.

For example, apricot and whiskey do not share flavors with each other but do have flavors in common with tomato. This creates a flavor chain that links all three ingredients, making them suitable to be used in the same recipe.

The team call this food bridging. They define it as “the ability to connect a pair of ingredients, that may or may not have a direct connection, through a path of non-repeating ingredients.”

This has an important impact on recipes. While food pairing intensifies flavor by mixing ingredients in a recipe with similar chemical compounds, food bridging smooths any contrast between ingredients, say Simas and co.

So what role does food bridging play in recipes from different cultures? To find out, Simas and co examined the flavor networks of cuisines from various parts of world and then analyzed the respective roles of food pairing and food bridging in each cuisine.

In Latin America, for example, recipes exploit both food pairing and food bridging, while East Asian food seems to avoid both principles. Southeast Asian cuisines such as Thai and Vietnamese seem to rely only on food bridging, while North American and Western European food use only food pairing.  

That’s interesting work that extends the principles behind the way we create recipes. Indeed, it reveals that food pairing is really a special case of food bridging in which the number of nodes in the flavor chain is 0.

A better understanding of these principles should help chefs create new recipes in specific styles. But it is by no means the be-all and end-all of cooking. Successful recipes have a wide range of different parameters in addition to flavor. There is the texture of the food, its temperature, its mouth feel, and its color, to name just a few.

Food bridging can certainly help with new recipes. But a truly universal tool for recipe creation will need to be much broader to incorporate these other factors into its model. That will require significant work.

But step by step, food scientists are learning how humans prune the list of all possible combinations of food to produce the combinations we actually end up eating.

Ref: arxiv.org/abs/1704.03330 : Food-Bridging: A New Network Construction To Unveil The Principles Of Cooking

Deep Dive

Artificial intelligence

Large language models can do jaw-dropping things. But nobody knows exactly why.

And that's a problem. Figuring it out is one of the biggest scientific puzzles of our time and a crucial step towards controlling more powerful future models.

OpenAI teases an amazing new generative video model called Sora

The firm is sharing Sora with a small group of safety testers but the rest of us will have to wait to learn more.

Google’s Gemini is now in everything. Here’s how you can try it out.

Gmail, Docs, and more will now come with Gemini baked in. But Europeans will have to wait before they can download the app.

Google DeepMind’s new generative model makes Super Mario–like games from scratch

Genie learns how to control games by watching hours and hours of video. It could help train next-gen robots too.

Stay connected

Illustration by Rose Wong

Get the latest updates from
MIT Technology Review

Discover special offers, top stories, upcoming events, and more.

Thank you for submitting your email!

Explore more newsletters

It looks like something went wrong.

We’re having trouble saving your preferences. Try refreshing this page and updating them one more time. If you continue to get this message, reach out to us at customer-service@technologyreview.com with a list of newsletters you’d like to receive.