Humans have evolved over millions of years as one of the most dominant species on the face of the earth. Over time, as we acquired enhanced cognitive abilities, we have also ended up developing a lifestyle that makes us vulnerable to diseases such as obesity, diabetes and cardiovascular disorders. While we may blame our genes, our social circles and sedentary work profiles for the rise of this epidemic of lifestyle disorders, diet is an important factor contributing towards these health issues.
 |
| Diet is central to the rise of lifestyle disorders such as obesity, diabetes, cardiovascular disorders and some forms of cancers. |
Cooking is a uniquely human endeavor which is suggested to be responsible for evolution of big brains in humans. Ironically, food is also central to many modern health problems. Experts have attempted to associate positive and negative effects of food on human health, without much convergence. The interaction between our body and food, leading to health consequences, is way too complex, giving rise to inconclusive and often contradictory assertions.
 |
| The interaction of food with body mechanisms is a complex phenomenon, giving rise to contradictory assertions about benefits and harms associated to food. |
I believe that taking a data-centric and evidence-driven view of food is the key to leveraging food for better health. With this idea, I would like to present our investigations of Indian cuisine in search of patterns and future directions for personalized dietary recommendations. Such data-driven studies are on opening new avenues for using food as medicine.
We started our studies by asking a simple question, “Why do we eat what we eat?”. What we eat on a day-to-day basis is dictated by traditional dietary practices crystallized as elaborate cooking procedures: the recipes. This question then gets transformed into, “Why we combine ingredients in our recipes the way we do?”.
 |
| The Question |
One of the possible answers to this question is known as the ‘food pairing principle’: ingredients which taste similar tend to be used together in traditional recipes. This implies that the traditional recipes have evolved to combine ingredients that are uniform in taste.
To investigate the food pairing pattern in Indian cuisine, we extracted data of traditional recipes from across different regions of India. These data comprised of more than 2500 recipes that are composed of around 200 ingredients from different categories: vegetables, herbs and spices, plants, nuts and dairy products etc.
Ingredients get selected to be used in recipes based on their flavor. And, the ‘flavor’ of ingredients arises primarily from how we taste and smell it, through what are known as gustatory and olfactory sensory mechanisms that are triggered by the flavor molecules. The pungency of onion and spiciness of chilles is due their flavor profile. So, we extracted the information of flavor molecules found in each of the ingredients used in Indian recipes, using various offline and online resources. Thus, each ingredient is now represented by a bunch of flavor molecules that characterize its unique taste and smell.
Having obtained data of recipes, ingredients and their flavor profiles, food pairing now is a measurable quantity. Each of the traditional Indian recipes was dissected into its constituent ingredients, to compute its food pairing. The average number of flavor molecules among all pairs of ingredients in the recipe. This number represents ‘the extent of flavor profile overlap among all ingredient pairs in a recipe’. When averaged over all the recipes, this number quantifies average food pairing across the whole cuisine.
 |
| Measuring "food pairing" in a recipe and cuisine. |
Food pairing is an objective measure that captures the molecular essence, the intuitive uniqueness of a cuisine. Similar to variations in regional languages, cultures across the world have evolved variations in the way they cook. Variations in the way they combine ingredients to form recipes, the unique mold that characterizes a cuisine. In the absence of cultural, climatic and other influences, the recipes would have been combined in a random fashion to create a 'Random Cuisine'.
Consistent with the food pairing hypothesis, it has been shown that many Western cuisines, such as North American, Latin American, Eastern and Southern European cuisines, indeed are characterized with ‘uniform food pairing'. These cuisines tend to blend ingredients that are similar in their taste and smell. On the contrary, studies from our lab have shown that Indian cuisine is characterized with ‘contrasting food pairing’.
 |
Indian cuisine is characterized with a characteristic 'contrasting food pairing'. More the extent of flavor profile sharing for a pair of ingredients, lesser is their co-occurrence.
|
This essentially means that Indian recipes tend to pair ingredients that have distinct molecular character. And, this probably could be one of the reasons for their unique taste. We found that contrasting food pairing is a general phenomenon across all regional cuisines. A quintessential feature of Indian recipes. It seems, despite diverse culinary styles, there is an underlying similarity across regional cuisines of India.
Across ingredient pairs, the more the similarity between two ingredients, the less frequently they tend to be used in the Indian recipes. Notice that the pattern in a Western cuisine would be ‘completely reverse’. More the flavor sharing between any two ingredients, the lesser is their co-occurrence.
We wanted to find contribution of each ingredient category towards the observed food pairing phenomena. For this, we randomized the recipes-- one category at a time. For example, to find how important a specific vegetable in recipes is, we randomly shuffled every vegetable with any one of the vegetables from the basket of ‘all vegetables available’. We found that such random shuffling affects food pairing only marginally for most categories.
Except for one: Spice. Random shuffling of spices in recipes with other spices disturbs the food pairing pattern significantly. This suggests that spice form the ‘molecular fulcrum’ of the Indian cuisine. Chefs suggest that such unique positioning of spices is in fact critical for the taste of a recipe.
Going further, we quantified ingredients for their contribution towards increasing or decreasing the food pairing. Among the top ingredients that make significant contribution to the molecular contrast, majority are spices: cayenne (chillies), capsicum, ginger, garlic, coriander, tamarind, clove, cinnamon and spice combinations (such as garam masala). These key spices provide the basis of food pairing in Indian cuisine.
While food pairing is a simple measure of molecular combination in recipes, I am tempted to link it to the taste. I must warn though that sensation of taste is a complex phenomenon involving a myriad of interlinked molecular mechanisms, and hence this suggestion needs to be taken with a ‘pinch of salt’.
Our data-driven discovery of this unique contrasting food pairing has been adjudged as an ‘Emerging Technology’ by the MIT Technology Review. Like knowing the law of gravity has allowed us to predict eclipses and to launch satellites into the space, I believe that such data-driven investigations of food will take us closer to developing divergent applications for food, nutrition and health.
With the variety of ingredients available, the number of possible recipes is astronomically large. Knowing the ‘culinary fingerprints of a cuisine’ can facilitate us in generating novel recipes that are hopefully palatable. Formulation of new food, food-beverage pairing, testing a food hypothesis, study of food-genome interactions, and mining food-disease associations, are among few interesting dimensions emerging out of our discovery.
One of the most exciting directions from data-driven and evidence-based investigations of food is that of ‘personalized nutrition’. In a pioneering study, researchers meticulously collected data of personal features such as nature of gut microbes, blood reports, body measures and food habits, from a large number of people. One of their meal was substituted with a standardized diet. These were then correlated with post-meal glucose levels, using a machine learning algorithm.
Interestingly, such a ‘personalized nutrition predictor’ could predict the expected rise in glucose levels even for a new set of people with a fairly good accuracy. More importantly, it could also suggest a personalized dietary recommendation that was used to successfully mitigate the levels of glucose, which is closely linked to Type 2 Diabetes. Now that's a big step towards finding solutions for diet-linked diseases.
 |
| Using data-driven investigations of food towards personalized nutrition. Adapted and simplified from Zeevi et al. |
This may sound like science fiction. But, who believed in weather predictions a few decades back. Despite the weather being a non-linear phenomenon, availability of large amount of climate data, along with computational and mathematical techniques, has transformed meteorology into a believable science today; at least for short term predictions.
I believe that the day is not too far when we will be able to find diet-based interventions for many life style disorders and leverage food for better health.
As a teenager, I grew up as an aspiring astronomer and astrophysicist. I saw this quote, displayed in the canteen of IUCAA, the Inter-University Centre for Astronomy and Astrophysics in Pune, while working on my master’s thesis. “The discovery of a new dish confers more happiness on humanity than the discovery of a new star.”
While I have not been able to discover a new star, with my data-driven explorations of food, I hope to be able to discover new dishes… Making humanity happier, and hopefully, healthier!
References:
[1] A Jain, N K Rakhi, G Bagler*, “Spices form the basis of food pairing in Indian cuisine”, arXiv:1502:03815 (2015).
[2] A Jain, N K Rakhi, G Bagler*, “Analysis of food pairing in regional cuisines ofIndia”, PLoS ONE, 10(10): e0139539 (2015).
[3] A Jain and G Bagler*, “Culinary evolution models for Indian cuisines”, arXiv:150500155v1,2015.
[4] Zeevi et al., “Personalized nutrition by prediction of glycemic response”, Cell, 163, 1079(2015).
[5] ED Sonnenburg and JL Sonnenburg, "Nutrition: A personal forecast." Nature, 528, 484 (2015).
Note and Acknowledgements: This blog has emerged out of talks delivered at various places in last two years, and is a run up to my TEDxDAIICT talk. Based on previous talks: Research Conclave, IIT Guwahati; HasGeek-KilterCon; Keynote Talk, iHOST 2017 at Le Cordon Bleu School of Hospitality, G D Goenka University; Cadence Advanced Technology Talk; Round Table on Great Indian Cuisine; CSIR-Centre for Cellular and Molecular Biology (CCMB); SERC School on NLD at Manipur University; School of Computational and Integrative Sciences, Jawaharlal Nehru University, Delhi; Shiv Nadar University; Central University of Rajasthan; Guru Nanak Dev University,Amritsar. I thank Mr. Shaayaan Shaikh, the TEDxDAIICT moderator, for his inputs on improving visuals.
