 So welcome everybody to today's session on an introduction to food science. I'm extremely grateful to two of our alumni, Ganesh Bangal and Huda Masood, both of who actually spoke at the first picture conference in 2017. I think if you were to think about a time has a very strange connotation at one point it seems like it would fly away. At another point, it seems like it doesn't pass. And I think for us, it has become a killer time just passes really quite quickly because we can't be there something or the other happening. But in this case, it's really nice that about three years later we've sort of like, you know, gotten into the loop again and both Professor Ganesh and Huda are here again, both of who spoke at kill. Professor Ganesh's work has been very fascinating and I will let him speak more about it and we'll let Huda make an introduction to it. But a quick introduction to Kilda and to our moderator for today. Kilda Club actually started in 2017 as a way to as a way to gather together as a group to understand the body to understand food and to look at all aspects of geekery around nutrition around health and science that goes with it. We had an interesting experiment in terms of people writing about their own experiments and their own journeys with food, with nutrition, with health, etc. And in the course we sort of discovered that, you know, I mean different people have different journeys. Huda has been running Huda Bals and I will let her introduce and talk about it more. I highly recommend that you try one of the Huda Bals. It's a better bet than Snickers. I'm sure Huda will cross the shoe at me if I want to draw that comparison. But Huda has also been delving a lot into the science of food into gastronomy and a few weeks down the line you will also hear from her about a topic that she really liked to begin to which is reading food labels. So with that, I'll hand it over to Huda and to Ganesh and I will mute myself to not have any more interruptions. Enjoy yourselves in today's session. Thank you, Zainab. Welcome everybody. Professor Ganesh Bagler is a professor at Triple ID Delhi and he runs a lab called Complex Systems. He's known for his pioneering research in computational gastronomy. This is an interdisciplinary science that blends food with data science. He is going to talk about computational gastronomy in the context of Indian cuisine, but it's not restricted to Indian cuisine. As usual, please put your phones off or put your phones on silent for the session. And if you have questions, there is a tab at the bottom of your screen which says Q&A. So if you can put your questions over there, please reserve your questions for after Professor Ganesh's talk, but make sure that you make notes so that you don't miss out. And yeah, we'll take it forward from there. Professor Ganesh. Thank you, Huda. Very good evening to all of you and let us get started with today's session. So I must thank Khazgeek and Kilter for giving me this opportunity for addressing this crowd about the food, the science of food and specifically computational gastronomy. So I'll be talking about computational gastronomy, about what computational gastronomy entails. It's something that I'll explain before I get started off. But before that, I'm Ganesh Bagler and I come from an institute called Tripoliity Delhi, Indra Prasth Institute of Information Technology. And those of you who have done any amount of binary number system would know that Tripoliity Delhi is 10 years old roughly. It's now 13 years running. And it has been dwelling on different aspects of computer science and allied areas which includes computational biology, design, applied mathematics and similar areas which overlap with computer science. My lab complex systems laboratory looks at various complex entities which primarily come from biological domain. And these entities are such as those of protein structures, cells, which basically we look at diseases as cellular, malfunctioning of cellular mechanisms, then brain as a network, we look at that. And finally, we have been looking at food as a complex system. So these systems, all of them are made up of interconnected entities, first of all, and they all exhibit an emergent property. Some of the parts is not the same as the whole. This is a telltale feature of all of these entities. Given the fact that today I'm known for computational gastronomy research, you might wonder whether I have been a foodie all the time. But I must tell you that as a teenager, I was an aspiring astronomer. So it has been a journey from being trained in physics to all the way up to in computational techniques to doing PhD in an interdisciplinary area called computational biology from Center for Cellular and Molecular Biology, Hyderabad to a couple of postdocs, one in NCBS, National Center for Biological Sciences, Bangalore to Max Planck Institution to all the way up to the discovery of this area or invention of this area called computational gastronomy. So it's been not so shortest path from astronomy to gastronomy. Having said that, what is computational gastronomy? My definition of computational gastronomy is that something that is a data science, which blends food with data and power of computation, so as to bring about data driven food innovations. So essentially, my bet is that like it has happened in many other areas of our life, where a certain aspect of life gets transformed by virtue of bringing in data and computation. Think of Google Maps, think of railway reservation systems and social networks and whatnot. All of these areas have got dramatically changed because we have been able to compile a large amount of data on different attributes and have been able to apply various algorithms, computation on it to be able to come up with solutions. Food remains an artistic endeavor. Food remains personal. Food is like my mother's recipe. It has got a very personal touch. So question is, such a personal topic, can it be transformed into a science in the first place and that to a data science, such that we can actually bring about innovations there? Well, all of my talk is going to be about that only, to pitch that indeed it is possible. And what is also not possible, I'll mention that in short run. When I look at food, I look at it as transformation of raw ingredients into delicious recipes. All of us are used to and we consume recipes on a day to day basis. So each of this recipe is a product of cultural evolution and has evolved over a period of a few hundred years or maybe more than that. And this food is magical. It's transformational. And incidentally, the ability of cooking is something rather unique to human beings, which is not found in any other species on the face of the earth. No other species on the face of the earth cooks its foods, relishes its foods in its cooked form in the delicious forms, right, and passes it on to from generation to generation. And human beings do that. And it has been argued that the disproportionately large brain sizes that we have is essentially is because one of the primary reason is because we got enough time from foraging because we could actually cook our food and obtain the necessary nourishment from the cooked food. To an extent, Richard Rangham argues in his book Catching Fire that cooking is the very essence of being human. Well, having said that if we have evolved to this extent, from primitive species to what we are today homo sapiens, you know, ironically, we have reached a stage where we seem to be plagued with an epidemic of lifestyle disorders, lifestyle disorders such as obesity, type two diabetes, cardiovascular disorders and similar disorders. Why many of these disorders could be factored in after accounting for various other elements such as the genetics and social factors and lifestyle that of sedentary lifestyle, etc. Food is a key component behind the epidemic of lifestyle disorders. And food has changed a lot over a period of time. Notice that maybe around 12 to 15,000 years back, the earliest of the food might have been or the cooked food might have been only roasting of a meat or a food for that matter. So over time, invention of instruments, invention of refrigeration, agricultural biotechnology, breeding, and many other techniques have changed the shape of the food the way human beings have been seen for a long period of time. My, the question that I'm asking standing today on 21st in 21st century is how is food going to change now, henceforth having seen what we have seen so far. The major, as I have explained to you a while back is that introduction of large amount of data associated with food, food attributes, along with the power of analytics, such as that of statistics, machine learning and artificial intelligence is going to change the way we look at food dramatically. That's my major. And accordingly, I have been spending a huge amount of time in my lab. In fact, practically, I'm completely devoted to developing the foundations of computational gastronomy for the last five years, and we have been developing algorithms on various aspects of food. And those aspects include those of traditional recipes because let us not forget that regard, whether we like it or not, all of us human beings, we have been locked into the cultural rubric of recipes and cuisines that we have been evolving over a period of time. So we need to understand that lock in, in terms of understanding the patterns that are present in traditional recipes, we need to look at the physical basis the flavors, which is what makes them possible to be included in a recipe used or not used, and the nutrition as well as health aspect of food. All of these factors is something that we are building both in terms of algorithms and databases. All of this journey that I've been explaining to you actually started with a very simple question. And that question that we asked when when I was still as at IIT Jodhpur was, why do we eat what we eat. Today morning I ate Dhokla. So why is it that I ended up eating Dhokla and what makes it possible. So the proximate answers are easy to be given here. Because the ingredients were available to me and you know that is what I could think of eating today morning. But the ultimate answer in terms of what make made Dhokla to be available in the form it is today is something which is far more deeper. So we took a very simplistic stance and we said that let's assume that a recipe is characterized by the fact that it is a set of ingredients as opposed to being individual ingredients. I know it's a very simplistic notion but let's stick to it for the moment and I'll come back to this point again. So by considering the recipe as a combination of ingredient this question changes into this one. Why do we combine ingredients the way we do in our recipes. Well, turns out that the you know one may have an intuitional answer to such a question. And one of them was coming from a chef called Heston Dumenthal from United Kingdom. He had this hypothesis of food pairing and he suggested that ingredient that tastes similar tend to go well with each other. And that's why in our traditional recipes, more frequently you end up finding ingredients that are tasting similar is what he proposed and in fact he used for creating such combinations. You know, the white chocolate and caviar combination chocolate and blue cheese combination, etc, which are very, very similar in terms of their taste and flavor molecules composition. You need to look at both aspect taste something which is perceived by the human beings when you consume it, as well as the other aspect is that of the objective aspect what flavor molecules are present in them. Now this is an interesting hypothesis and in print. In fact, this hypothesis was tasted in 2011 by looking at what can be called as Western cuisines. I know that this is calling a cuisine as a Western or Eastern or Indian is a is a rather tricky matter. We need to look at the factor social factors anthropological factors, etc that gone that have gone into the creation of a cuisine. But nonetheless, if I were to bundle all the recipes that are part of some of these western some of these countries and continent such as North America, Latin America, Southern and Eastern Europe, Europe, etc. Then I can call them Western cuisines. So this particular principle has been shown to be true to in the Western cuisine. What we did was to ask this question in the context of Indian cuisine, which is diverse culture rich and has health centered dietary practices. And this question wasn't dealt with in the context of Indian cuisine. So we started looking at the data of Indian cuisine. Now, notice all of a sudden food recipes cuisines which were otherwise not so scientific in a scientific domain, they were more like an artistic endeavor, they are now turning up into a data endeavor. So we collected the data of recipes of Indian recipes from across various parts of India. And this was the earliest data set that we came up with in 2014-15. And that included data coming from tarla.com, which had around 2500 odd recipes, which were belonging to regional cuisines which were across length and the breadth of the country. And each of there were around 200 odd ingredients to be specific 193 ingredients, which belong to one of these food categories that of dairy, fish, fruit, herb, meat, pulse, vegetable, spices, etc. Having got this data of recipes and the ingredients of which these recipes are composed of, the next thing that we did was to look at the flavor basis of these ingredients. Because the onions and the chilies and the tomatoes and the tamarinds would be chosen or not chosen to be a part of a recipe by virtue of what is their flavor, how do they taste, how do they smell, etc. What is their aroma, right? And incidentally, flavor molecules which are part of these ingredients, the chemicals are the ones which when we consume this food or when we take it closer to the nose, interact with the so-called olfactory and gastatory mechanisms in our body, giving rise to a rich sense of the flavor of that particular ingredient, be it onion or chili or tomato. Given that we compiled from a large body of literature, the flavor molecules that are reported in each of the ingredient. So essentially, corresponding to onion, we had a bunch of molecules which are known to be found, flavor molecules, which are known to be found in onions, and the same was done for each of the ingredient that we had with us. Having created in this manner, this is the broad statistics that most of them tend to have relatively small number of molecules, you know, anywhere between 0 to 25. But there are certain ingredients which do tend to have up to 50, 75 or maybe even 100 ingredients, 100 flavor molecules in them, right? So this is the statistics of that suggesting that some of the ingredients are pretty rich in terms of their flavor profile. And having got the flavor profile, most importantly for us, we can do the study that was suggested by the food pairing principle. And that is to look at what is the similarity of the flavor profile between any two ingredients, like in this case of coconut and onion, right? So how do we do that in the context of a recipe? So we looked at the full cuisine, the Indian cuisine itself, starting with recipe number one, recipe number two, all the way up to 2543, right? And we broke down this caricature coconut chutney recipe. I deliberately say it's caricature because obviously it's not real recipe, where let us say it has got this coconut, onion, chili and curry leaf. We can break it down into its constituent ingredients as shown in the bottom of the slide. And further, we can actually create pair wise combination of these ingredients like onion and coconut, coconut and chili, curry leaf and chili, etc. And for each of these pair, we can ask this question about what is the number of common molecules between the two, right? If you do that, then you will end up getting one number for each of these pair about how many molecules are common if there is any. Now, when you take an average of all of these numbers that you get, that comes out to be the food pairing index of that recipe. For example, it could be that the food pairing index of coconut chutney is let's say 5.31, which is an average number of flavor molecules shared across all pair of ingredients in its ingredients. Well, the beautiful thing that we have ended up doing here that with so many... Sorry to interrupt over here. In its recipe, correct? In regards to the curry leaf and the onion over here. True. Can you explain a little bit more about what kind of pairing is being done between something like say coconut and onion? Just so I have a clear idea. Yeah. So I don't have the exact numbers with me, but let us say the coconut and onion have approximately 15 flavor molecules shared between the two. Okay. And similarly, coconut and chili may have three molecules shared between the two. These are molecules which eventually going to elicit taste and order when we consume them. And when you take an average of that, you will get a single number, which is the average number of flavor molecules. I hope it is clear. And the higher the food pairing is, the more complimentary the ingredients are to each other. Higher the food pairing is, more similar are the two ingredients to each other. That is true. Yeah. Please go on. Yeah. So having got the food pairing index in such a manner for one recipe, you can actually do the statistics for the whole cuisine. So remember this was one particular recipe for which we did this computation of the food pairing index in terms of flavor molecules is sharing. Right. And that number which comes out from the whole cuisine is the mean value of the food pairing across the whole cuisine and its standard deviation. All right. So this is done for the statistical purpose. Why are we doing this? We want to ask a question that given the basket of same ingredients, same vegetables, same dairy products, same chili, same spices and herbs, etc. The question is, if I were to randomly put together the ingredients, what would be the food pairing index because even such a randomized bunch of ingredients will have some amount of sharing amongst them. So compared to such random cuisine, what would be the food pairing index of a real cuisine? That would tell me how uniquely the cultural practices, the geography, climate, etc. have affected the cuisine to be what it to become what it is today. So we ended up doing this experiment to find out what is the food pairing index of a randomly composed recipes, ingredients being exactly the same that they were in the beginning. So when we do this experiment, turns out that compared to the randomized cuisine, the Indian cuisine, the Western cuisine, tend to have a uniform food pairing, which means they tend to have ingredients that are very, very similar to each other in terms of their food pairing, which would mean that there is a overlap is pretty much higher here in the ingredients on an average, which are found in Western recipes, and that's what we have labeled it as uniform food pairing, similar type of it. So compared to that, when we looked at the Indian recipes that we had with us, we figured out that the pairing is in fact of a contrasting nature, meaning compared to that of a randomized bunch of it, we tend to have lesser and lesser overlap, lesser and lesser molecules that are shared between two ingredients in on an average in Indian recipes. And that is what we named it as contrasting food pairing. And this turned out to be an interesting observation in the context of recipes from Western world. Again, I must point out to the fact that we are stereotyping here, but nonetheless I'm going to use the notion, broadly speaking, the Western world versus that of the Indian recipes, India itself is pretty diverse but we have taken all of these recipes together from across the length and the breadth of the country. Well, one of the most interesting observation that we did was actually the second one, wherein we did random shuffling of ingredients within a given category. So we had bunch of recipes which constituted a cuisine, in this case Indian cuisine. What we did was to say that look, what would happen if I were to take every vegetable in these recipes out and randomly replace it with some vegetable from the vegetable basket, everything else being the same, everything else is untouched, absolutely. When you do this experiment, turns out that you will have some change in food pairing. And that would indicate how sensitive is food pairing to such shuffling within vegetables. Right, so that's the experiment we did for every category that we could manage to do based on the statistical numbers that we had with us. And we figure out that compared to the original food pairing that the Indian cuisine had, the contrasting one, as you can see, the shuffling within category for vegetables or fruits or dairy or plant, etc. was changing the food pairing marginally and still keeping it on the contrasting side. Whereas, if I were to shuffle one category of ingredients randomly, the food pairing index was dramatically getting changed. And you can actually make a guess about what that category would be. And most of you would be right about it. The reason being that it's pretty intuitional, I would say, about which category, ingredients of which category are critical in Indian recipes, spices. We found out that spices are the molecular fulcrum of Indian recipes, meaning that shuffling them, taking a clove out and putting cardamom or taking cardamom out and putting something else there is actually going to change the pattern of food pairing pattern, and thereby probably taste and the aroma of the recipe in a manner that it changes the identity of the cuisine. So we concluded by the end of this study, I must also highlight that while the contrasting food pairing result is very interesting. The fact that spices are the molecular fulcrum of Indian cuisine has remained a robust result, despite the fact that we have collated a much larger number of recipes than we had earlier. Despite that the result has remained pretty robust. So this is a very, very strong result and something which doesn't change with change in data which has happened because of the very early stages of this particular area. Professor Ganesh, if I could just like go back to the swapping out of specific ingredients. You had mentioned that when you swapped out vegetables there was much lesser change in the taste profile. In the food pairing index, how it reflects in taste profile, I can't say that it's difficult, I don't know that. Okay, but the food pairing index does denote the molecular olfactory and gustatory molecules that are actually going to play a role. So we can extrapolate from it a little bit. Yeah, it's a complex function of molecular composition, but yes, that's a guess we can make. I'm deliberately not making that guess because it would be a long hall, would be making a jump from there. Okay, and then you're stating that the biggest change that comes in the food pairing index which just throws all the numbers off is when you swap out spices instead of ingredients like vegetables. Maybe in like a sub-Z. Okay, all right, thank you. Yeah, that's observation. All right, so what we did is to notice that all of these observations of patterns in the recipes data are very myopic. They look at simplistic patterns such as pair wise ingredients, and they're rather naive. So what we did is to create a larger repository of patterns that can be measured in a given recipe or in a given cuisine, not recipe in a given cuisine to look at which category of ingredients are dominant, which ingredients are far more frequent. What are the pair wise combinations which are more frequent, which are the triads three ingredients put together which are more frequent and higher orders of combinations, etc. All of these things put together, we labeled it as a culinary fingerprint, like a DNA fingerprint which uniquely identifies a human based on the patterns in the DNA in a similar manner by looking at a large number of recipes that are known to characterize a cuisine. One can actually pull out these patterns by looking at their recipes, ingredients and the flavor molecules to form culinary fingerprints. That's what we ended up observing and finding it out and publish it later. Going forward, this culinary fingerprints actually tell us about the role of a particular category of ingredient like spices, vegetables, dairy, dairy products, etc. How do they contribute towards the food pairing index, that is what we measured, and this is just an illustration of a South Indian cuisine, and how does it look. All of this research that we did in 2015 got a lot of press and also academic recognition. It was identified as best of 2015 and also highlighted as an emerging technology by MIT technology review and also many people such as Veer Sangvi and many other outlets across more than 200 media outlets highlighted our work as the future of the food. And importantly, industry was not far behind in recognizing the fact that bringing data and computation together with food is going to revolutionize food in a big way, although it will take its own sweet time to make it happen so they identified it as the sweet spot of the food. What I'm going to do now in the next 15-20 minutes is to tell you about what is the future that we are looking at, what is the future that we are building in our lab when it comes to computational gastronomy. And while saying so, I must also highlight what are the shortcomings of the work that we are doing as of now. Notice that computational gastronomy is where physics was maybe in 18th century. It's a very early stage. Food is a very, very subjective matter, very personal matter. It has got variety of associated factors, you know, geographical, climatic, personal interpersonal relationship and whatnot apart from the physiological basis of food that it may have. So therefore, unentangling all of these parts to bring out food in its full glory when it comes to computational gastronomy is going to take its own sweet time. And this is the story of a spherical cow. Those of you who have not heard about it, physicists and mathematicians are known to make this mistake of when given a challenge of modeling a cow, they will start with the assumption that let's assume that the cow is spherical. Well, turns out that there's such assumptions are useful at times, but at the same time, they would not be representing the reality in the best possible manner. Therefore, the kind of studies that we have done, while they are useful, interesting, give us insight into a social anthropological factor such as food. At the same time, it doesn't actually let us to the depth of food at the level at which we would like to go. And therefore, let me tell you what are the assumptions that we made. We said that recipe is an unordered list of ingredients. Of course not. Recipe is an ordered list of ingredients and quantity matters the process which goes into cooking also matters. Similarly, we also assume that an ingredient such as tomato or a chili is nothing but a bunch of flavor molecules practically speaking. But in reality, that is not the case. The concentration or the quantity of each of the flavor molecule that goes into the ingredient matters or how that molecule is interacting with the taste and like order receptors matters a lot. Right. So there are many such factors which have gone ignored. And I must tell you that we are trying to build upon that whatever is level possible. And it's a new niche that has been created. Therefore, the number of people who are working and the number of people are willing to spend their time dedicate their time in building the foundations are relatively few. And I'm proud of it that our lab is trying to do this job of putting things together of this piece of puzzle. What we are trying to do in our lab is to bring about data driven food innovations by using this data centric and AI centric platform. And there are three dimensions practically mainly speaking that we are looking at although there are other dimensions that not going to talk about today. One is that of the recipes that how can we dramatically change the way we look at recipes. Second is that of nutrition and health. How can we make food more tasty more nutritious more healthy. And how can we change food at the level of molecules is these are the three different angles that I'll be telling you in the next 15 minutes or so before I can conclude. So the first angle was that recipes had been looked at in a very simplistic manner. So what we have done going ahead is to compile the recipes not only in a couple of thousand but in 100 thousands. So we have now this repository of 118,000 plus recipes, which is called recipe DB now getting published in a journal called database, wherein we have compiled the recipes from across the world. And that includes 24 world regions and 75 countries in a computable format. Unlike a traditional recipe resource where you can go and search by name of a recipe here you can search by complex queries. Show me all recipes which are having tomato chili but doesn't have turmeric from Algeria, Algerian cuisine, and such questions can be asked which can potentially be answered by a recipe DB provided the data is available there. And we have of course a lot of attributes, the type of recipes, the type of cooking processes which are being used in the recipes such as frying, boiling, sauteing. So we have such 160 odd processes which going to it starting with very simple such as add stir to all the way up to more complicated ones. And importantly, we have broken down the recipe to the level of its quantity unit being used such as teaspoon spoon and similar and the kind of temperature state and size at which they are cooked. All of these have been also mapped to their nutritional data from United States Department of Agriculture nutritional data to estimate their nutritional profile. The agenda here was to come up with a system where given a recipe in a human readable format, it should be able to extract all relevant information and tell us about, especially its nutritional value, both micronutrients as well as micronutrients. That was the objective with which we worked on it. And this is not a job done completely. This is the step one the V1 is what we have created and we hope to build this further to make it far more applicable. We have gone further and looked at these recipes from various parts of the world, and have done culinary fingerprinting of these world cousins remember we had done it for Indian recipes Indian regional cousins. And we have gone further and done it for the world cousins to be able to map what is what are the patterns in these cousins across the world. One of the interesting question that we have been asking him in our lab is that, like languages have their own way of similarity in them, the phonemes that we use them, the letters which are used in written in language. Based on that we can construe the similarity of the languages. We know, for example, how Punjabi Urdu and Hindi are similar, and that way Telugu and Kannada are very similar, both in spoken as well as in written language. And thereby we can actually build a tree, which would have a Dravidian branch versus that of an Indo-Arabic branch etc. So the question is, can we also similarly learn from the patterns of the cousins without going into much of the details and come up with a tree of cousins. That was the idea behind it, imagining almost as if cousins have been evolving over a period of time, although there is a far more subtlety that goes into this notion because we are borrowing from biological evolution. But nonetheless, the primary idea is acceptable and we have built such cuisine tree similar to language tree. And this is the tree that I was showing to you. One of the questions that we are asking in our lab is, theoretically how many recipes are possible? And the number turns out to be astronomically large. The average number of a ingredient in a recipe is around 10. And the number of ingredients that are available to be put in these slots are around 1000 with a very, very conservative estimate. In reality, the number is in tens of thousands. But I'm taking a very small number, which is definitely agreeable. And with that you get a number which is 10 to the power 30, which is astronomically large. So the question here is, why is it that the cultures have evolved recipes which are far less in number at any given estimate, the number is far less than 10 to the power 30. Very, very, very, very less, minister. So question is, can we actually come up with strategies by which we can learn from the recipes that have been already evolved and build new recipes? Well, that's the question we are asking in our lab. Our inspiration is also coming from a movie called Ratatouille. I'm sure many of you would have seen this movie called Ratatouille, where Chef Gustav has this catchphrase where he says that anyone can cook. So the question here is that can the intuition, intelligence of a chef can be built into a program, an algorithm which can spew out recipes which are new recipes. Remember, chefs and culinary enthusiasts are known to tweak, do try an error and identify which recipes are good and come up with new recipes probably over a period of time. In fact, many people don't argue with the fact that you can build a knowledge repository of cuisines across the world like we have tried to build in recipe DB. But how will you build in the intuition? That's a big factor. And we are trying to do that using the deep learning algorithms that are in vogue right now. And they have been trying to replicate the style of writing of Shakespeare, the style of music synthesis of Beethoven. And we believe that we should be able to regenerate expert chefs intuition with the help of large number of recipes and in a structured manner. They are not just a bunch of words, but where we know this is an ingredient, this is a quantity, this is a unit and thereby can we build new recipes is the question that we are asking. So it's a very, very challenging and interesting dimension that we are working towards coming to the second dimension that of flavor molecules and the molecular space, wherein I would like to tell you that until now, as I said, it's a new area. Therefore, there was no existing repertoire of flavor molecules found in natural ingredients, those which are found in tomato ones which are found in chilies and etc. It was not available until now. We have built this repertoire called flavor DB, which he was also made available as a flow app. It has been taken down now for upgradation purpose, but this repertoire provides the most exhaustive collection of flavor molecules that are reported in natural ingredients. And the idea is also to come up with applications out of it very soon, while the agreement is still to be done, but very soon we may deal with a big company, which brings you into an voice answering system, the flavor DB app where you can do food pairing with the help of one of the applications of such a compilation of flavor molecules, where in molecules and their tastes are put together is that of predicting taste. Can you predict, given a molecule and its molecular structure, what taste it would be off? Would it be bitter, sweet or tasteless? We have built algorithms which have been now published state of the art algorithms, which can predict taste. And you would wonder what are the possible applications of this? Well, finding out novel molecules which are sweet but at the same time don't add colorific value is one of the most challenging problem in this diabetes-ridden world. And that's the challenge that hopefully our algorithm will help addressing towards. Eventually, we would like to build a food maker's guide to the galaxy. Synthetic molecules are nothing new to food and flavors and beverages industry, fragrances industry for that matter. What we are trying to suggest is this, that the number of organic compounds using carbon, hydrogen, nitrogen, oxygen and some stray elements that go into organic compounds can be built is also infinitely large. But can you do it in a manner that you can recreate a particular taste, odor, aroma? Can you do that? Well, that is what we are trying to build as part of the food maker's guide to the galaxy. The last angle that I would like to touch upon is the space of health, apart from dealing with recipe, nutrition and flavor molecules part. Well, always we come across contradictory assertions about how a particular food is good, but at the same time also bad for you through newspaper articles which are writing their report based on some scientific research published somewhere. Well, I believe that part of the reason why these assertions are made in a contradictory manner is because of the way the food interacts in a complex manner with the body, giving rise to health implications. Right. So this interaction is not so straightforward or simple as shown in this particular picture. And therefore there is a need to integrate all the data that is available about food ingredients and their health or disease associations. That's what we have done. We have dug out the data of PubMed and Medline Medical, the scientific repertaurs, where in scientific research is published about a particular ingredient being good or bad for you and have put together a repertoire of 2200 or food ingredients and their disease associations. Going further, we have also integrated these data with that of chemicals which are found in the food genes which are known to be present in the body which interact with the chemicals food chemicals, etc. So barring the food and gene interaction, every other data is empirically available and we have created this repository called DietRx and we are in the process of analyzing it to find out major signatures that come out of it. One of the utility of such a repertoire was to ask the question about why do we use spices in our food. So without getting into much of the detail, very briefly, the proximate answer to this question is because of the fragrance and flavors. In 1998, back to back papers Sherman and Billing had actually pointed out that spices and herbs, culinary herbs and spices tend to have antimicrobial properties beyond their fragrance and flavor that we know them for and therefore they may be used for that purpose. What we have shown in 2018 paper after 20 years of their work is that beyond microbial properties, the culinary herbs and spices also tend to have broad spectrum of benefits across cardiovascular disorders, immune system disorders, intestinal disorders, etc. They have beneficial effect and these conclusions have been drawn from a large number of research articles. There are 38000 odd research articles that we have in our repertoire. Well, where is it all moving towards to conclude? In my opinion, all of this research of computational gastronomy blending food and data with computation is taking us towards a world where food would be personalized. In the world of COVID that we live today, we know how much important it is for us to be having a strong immune system, which helps us cope up with viruses such as COVID-19. So, the question that one was asking in this particular paper that I'm showing you from Weizman Institution was, can you build a machine learning algorithm, a program, which by measuring a variety of body parameters would be able to club it with a particular food or a diet that a person will consume and will be able to predict. Finally, what is the post-prandial glucose level, the increase in glucose level in the body of that individual? Well, turns out that this particular paper is one of the landmark papers in the area of food and nutrition. And it actually showed us that indeed it is possible for us to come up with personalized nutritional predictors. And I believe personally that we are moving into that world where personalized nutritional predictors would be routine and would be using them all the time. Going beyond the broad predictions that we tend to make today. So in my lab, we have been building applications of various sorts, including food beverage pairing, prediction of taste and order, fingerprinting cuisines with culinary fingerprints, dietary intervention creations, beverage and food design, as well as sustainable food innovations using the computational resources and repositories that we have created, not to mention the algorithms that we have been building all the while. And this is also leading towards a startup called gastronomic acetic, which blends food with artificial intelligence and tries to come up with new inventions when it comes to food innovations. So I'm going to end this story to tell you what I told you in the beginning that I was an aspiring astronomer, while I could not become an astronomer, I ended up becoming a gastronomer, a computational gastronomer at that. And I remember this quote from Ayuka Inter-University Center for Astronomy and Astrophysics, where I was working as a master's thesis student that the discovery of a new dish confers more happiness and humanity than the discovery of a new star. So while I have not been able to discover new stars by being an astronomer, I believe, I hope, I wish I should be able to discover new dishes with the endeavors that I am right now doing with the computational gastronomy work. So with that, I would like to thank you for listening patiently, and I'm open for questions and answers. Just give me one minute when I could actually thank a variety of people who have helped me in this endeavor, wherein we are trying to put knowledge data and applications together in my lab. A large number of PhD students, MTECH students, BTECH students, youngsters who have helped me, who have been my collaborators, who have contributed to different aspects of the work that I've done, without which most of it would not have been possible. Only ideas would have been possible, but implementations wouldn't have been possible. Right, so I should thank them, all of them. And many of them have been summer interns, so we have a very active summer internship in my lab, which goes on every year. Right now also it is going on in online mode. So with that, thank you, Noot. I get back to you for question and answers. Thank you, Professor Ganesh. This is a very interesting and exciting look into computational gastronomy. I would like to start addressing questions to you, whatever we've collected over the course of your talk. Priyanshu asks whether culinary fingerprints performed only using the flavor molecules or other parts of food as well. Culinary fingerprints entail using information of recipes, ingredients, as well as the flavor molecules, not to mention even the functional group present in the flavor molecule. So it's a multi-layer information that was used for creating culinary fingerprints, Priyanshu. Thank you. Mr. Prasad asks, some doctors recently are speaking about human microbiomes and how post-COVID-19 world, they will be very significant. So how would computational gastronomy help in building up, say, good gut microbiomes? Thanks for asking this question. I obviously hadn't touched upon many aspects of food that included gut microbiome, which happens to be one of the hot topics these days in the food sciences. Quite clearly, in fact, the Weizmann Institution research that I pointed out towards the end, indeed uses gut microbiome as one of the factors which goes into the machine learning algorithm. So wherever you are compiling data and analyzing it, that becomes part of computational gastronomy. So as long as you are looking at gut microbiome data, plugging it with other data such as BMI, food diary, and what kind of diet that you are having, thereby making predictions of any kind of health indication. That way, computational gastronomy can pitch in there. That was also a machine learning algorithm. And similar kind of algorithms can be built for suggesting what kind of dietary interventions are beneficial against diseases such as inflammatory bowel diseases, cones disease, etc., which are diet linked. Thank you, Professor Ganesh. One more question. How do you think that the field of psychology, Humeirah Fatima asks this, how do you think that the field of psychology and neuroscience add value to computational gastronomy? Okay, so this is a tough one. Tough one, tough one. Though this has been thought about, this has been thought about and very complex one. Food is a complex topic. So definitely psychology touches upon it. With whom you are having your food, the kind of music or ambience in which you are having your food may make a huge difference in your perceptual, how you pursue food. This has been studied incidentally in the field of psychology and somebody suggested me from Tripoli to Hyderabad that I should collaborate and work on it actually. So I believe that indeed computational gastronomy, which is a broader paradigm, wherever you use computation in the context of food, it can be called as a computational gastronomy, largely speaking. So I believe there is a much scope for working on questions which are derived from psychology, not to mention the neuroscience. In fact, only next week I have a meeting with an entrepreneur who is working on neuro gastronomy about how food elicits different reactions in different parts of the brain. And can we analyze this in the context of different food molecules that the food has? That's one of the broad question one is asking here. I personally believe there is a huge scope. How soon this is going to happen? I wonder, will it take time? I believe it will take a little time before we can touch upon it. Among these I believe neuro gastronomy is something which will come up very fast, very soon. People would like to find out what are the neural correlates as they say of a certain perception. Thereby, if you would like to evoke that sensation, one of the immediate application that I can see here is this. Can you elicit somehow, quote unquote somehow, sweetness without having sugar in your food or your beverage. If you can do that, you have a million dollars or billion dollar solution probably. Right, so this is a multi-billion dollar solution. Thank you. The next question is, there is molecular gastronomy which is innovations used by chefs. How is computational gastronomy different from molecular gastronomy? And a second part of the question, if impairing the index is negative, does it mean that paired foods together are not healthy? Okay, second question. First, I'll take it right. So first of all, this positive and negative food pairing doesn't have anything to do with food being good or bad. It is simply a mathematical notion used on a vertical axis, something which is random, something which is on one side, the so-called uniform side or positive side, and something which is on the other side, which is contrasting side. That was the only reason the word negative food pairing was used in the research article that we wrote, that is to put a contrast, to create a contrast with the positive food pairing. But after realizing that many people are misreading it to be having negative health consequences, we are now sticking to the terminology of uniform pairing and contrasting food pairing. So that's the answer to the second question. I'm sorry, but I lost on the first one. I'm very quickly. Can you repeat that? What is the difference between molecular and computational gastronomy? Now I recall. So molecular gastronomy is primarily a chef's endeavor, whereas chef is trying to recreate a certain test using certain molecules, like you would like to create a taste of a recipe using only a few molecules. For example, that is what molecular gastronomy largely entails with. That's a very, very physical endeavor where we are trying to come put together molecules, which will, for example, it doesn't have palate, it doesn't have paneer, but it tastes like palate paneer, right? By putting together some molecules together, you would like to do that. That's called molecular gastronomy. So computational gastronomy is far different. It basically creates a framework, data-driven framework around food and food-related attributes, the one I told you, recipes, ingredients, the flavors, nutrition and health, and thereby try to create applications out of it. So it's very different from that. Mr. Hariharanasi asks, how would you advise using this research to develop new recipes for recipe developers? Okay, so there are two ways of doing it. Maybe you can talk about your favorite recipe, which is your bhaigan ka varta. Okay, so, okay, I have got a lot of feedback, including Chef Garima, who is a Michelin star chef actually herself, saying that the food pairing parameter that we have identified, she finds it useful while designing her recipes. So that's a chimeric job. You are using intuition, but at the same time, you are also trying to use objective parameters given by food pairing index. So we have FlavorDB, the database. You go to FlavorDB, there is something called food pairing, and the food pairing index actually provides you, tells you that if I ask what is similar with mango, then it will tell you all the ingredients that are similar by virtue of shared flavor molecules. Similarly, it will give it to spinach or tomato or chocolate, whatever you want. Now you can use your intuition that, oh, let me create a uniform food pairing, similar one or contrasting one by choosing ingredients of your choice, taking ingredients of your choice, putting them together and figuring out whether it works or not. I have, of course, many Chef friends and because I conduct a symposium every year, so we have already conducted three symposiums. Chefs come to that, they learn from us, they go back and do experimentation. Many have got back to us, not all, but many have got back to us saying that this tool helps them in short listing ingredients that could be good for creating such recipes. So that's a very loose answer for the first part, where you can use your intuition. Second, you'll have to wait for it. We are in the middle of creating novel recipe generation algorithms, which, for example, would learn from all the biryanis and will come up with a new biryani or 10 templates of biryani. And then you can choose one of them which can possibly be, in your opinion, the best recipe, alternative biryani recipe, right? But that will take a while. It's a deep learning, machine learning based algorithm that we're trying to design. It has to beat a human chef, without which we can't say it is a really good one. So we are in the middle of doing that. So it's a futuristic work. So Mr. Hariyan and you would have to wait for a while for this to come to reality. So Professor Ganesh, this is a lot like, you know, maybe building perfumes or maybe building very complex perfumes, right? You have a top note, you have a middle note, you have a bottom note. So if you're going to do like a template for biryani, you want to know what it smells like when you open the pot. You want to know what it tastes like, what immediately hits you, what kind of acidity hits you, and then the texture, the umami texture that you feel across the board. And yeah, we can't wait. The chefs that you were talking about, who attend your symposium, what tools do they use when they attend your symposium and go back to their own to experiment? Most of them used positive food pairing principle that seems to be still a dominating principle incidentally. So they essentially went to clever DB. They, for example, looked at chocolate and what are the ingredient that pair or have most similarity with chocolate. And while creating a chocolate based recipe, they use those ingredients as possibility. And wherever they found the exception, they actually got back to me and also told me that look, this is not working well, despite it being so much similar. It's not really working. My customers are not liking it. Just to give an example. Is this, is this database available to the general public? Correct. Flavor DB is available as a browser, browsable database for non-commercial purposes, as long as you use it for making some player playing around, etc. It is available for commercial purposes, a purpose that is not available just like that you will have to make a contract or something. Thank you. Now, Mr Priyanshu asked on what lines the field of gastrophysics and Professor Charles Spence's intersect with computational gastronomy. I didn't even know something like gastrophysics existed, but today I learned. Gastrophysics is more on lines with molecular gastronomy, I would say, pretty much, whereas computational gastronomy is heavy, very heavy duty on data and computation. I would rather collect data coming from Spence's lab and use it for doing some analytics at my end. That's computational gastronomy. If I were to create a lab to put together ingredients, molecules to create a desirable recipe or a flavor, that would be Spence's lab. That would be molecular gastronomy or gastrophysics. Gastronomy and gastrophysics are pretty much similar, except that physics principles are used a lot heavily when they are trying to design a new recipe, such as that they do in Spence's lab. Okay, I'm going to put you on the spot again. You very cleverly avoided my question with regards to your favorite recipe and how you've played around with it. Trust me, I still remain a very novice cook. I fail a lot. And while my wife takes some intuition from what I do, my mother hasn't done as much, but my wife has done it. But I am still playing and trying to figure out how do I change Megan Kabharta in a recipe which is more or at least equally likable by changing some of the ingredients by using food-pairing principles. But so far I have not been able to change my liking. What it is, primary liking is with the basic, the classic Megan Kabharta that hasn't changed with any of the food-pairing that I have done so far. Do you think this is something to do with nostalgia as well, because you've grown up eating it? Yeah, possibly. And it is something that I know that we've touched upon when we spoke earlier with regards to the flavor molecules, both our factory as well as the straight tree, evoking nostalgia and that also being therapeutic in some aspects. It is eminently possible. As I said, it's a complex topic. You know, those who have grown up, there is a factor of nostalgia and more like, there is another word for it. Basically, you have been used to that particular taste and order and aroma. Like somebody who has grown up eating Megan Kabharta done in Chulha would not like it as much on a gas probably, right? It's that kind of a difference it makes actually, right? So it's a complex topic. Food is very complex. What we are trying to do is to take out the top layers only. That's why I mentioned that it is probably still 18th century physics or 17th century physics. Very early stages, very basic principles is what we are trying to figure out from whatever we have available with us. That's wonderful. Now, I have one more question. When the chefs come back to you saying that this didn't work, was there one particular example that stood out? I can't recall that particular example, but it was regarding chocolate. With chocolate, one of the ingredients that was coming on the top in the third or fourth number in terms of chocolate pairing with something else. That's something else. I have a written record of that somewhere. The person wrote a message to me saying that this ingredient didn't really work well. So I do have that particular example, but I don't know the name of it right now. And there are many more. Sorry. And are you investigating why despite it being so high up on the flavor or the food pairing index, it's just not working well? I looked at it if there is an easy answer to it by looking at its flavor composition or some unusual molecule which might be present, which is giving an unpleasant odor or aroma, etc. But there was nothing so easy to come by. If you look at flavor DB, you will come to realize that it's a very, very complex database. It gives you molecules and molecules' order, what kind of order or a taste the ingredient molecule has. So it is very difficult to pin down in the absence of additional data, such as concentration of the molecule, which is not available. So in chocolate, for example, this molecule, what concentration it is available in? Practically, it is not available for every ingredient. We have around 1000 odd ingredients in our database. So in the absence of large amount of such data, it is impossible to investigate and pin down about what made that possible. Thank you. Mr. Hariharan has another question for you. I think this is similar or it's just writing off of what you've just said right now. Do the flavor molecules of ingredients also change based on the method of cooking? If yes, how is it accounted for in the flavor DB? Good question. A very good question. So, of course, flavor molecules change their composition, their nature when they are processed, depending on whether they are boiled or fried or sauteed or they are raw. They are obviously very different from each other. In flavor DB, flavor DB is not cooking related database. Its flavor molecules is repository. So natural molecules as available in raw ingredients is what is reported in flavor DB. If you ask me for boiled beef and fried beef, boiled chicken, for some of these ingredients, we do have their molecules available with us. But for most of them, remaining thousand-odd ingredients, we don't have their transformed form. In fact, one of the major agenda, if one were to do some interesting gastronomical project is to do this, is to take an ingredient, put it through some popular protocols that it goes through, boiling, frying, sauteing, roasting, etc. And look at the flavor composition using gas chromatography kind of techniques. What was the original flavor? What are the flavors later? And make a list of those and create a database of that. That itself would do wonders to this area of computational gastronomy, propelling it towards the future, trying to create new recipes out of it, actually. In the absence of that, we are heavily limited. So to answer the question of Mr. Hariharan, we don't have that data in flavor DB available with us. Not right now, but I'm sure you'll have in terms of excited to put chicken or mutton or beef through the entire processes that you said, whether it's boiling or braising or sauteing or even to weed for that matter, or deep frying and then get the database, get the molecules that are required and then put them into the database. Yeah, yeah, we intend to do that. And one of the, you know, lackunas of shortcomings of being in an area such as computational gastronomy as an active researcher, I can tell you, is that if you work on cancer or diabetes, you will get million of funders, whether it is from DST, Department of Science and Technology, Department of Biotechnology, etc. But when you're working on food and that and you develop a new area as a pioneer of computational gastronomy, everybody claps saying that wonderful. You're done amazing thing. Nobody would have done this. You have spent five years so far working on this area, etc. But when it comes to funds, practically none. So this needs a lot of money to be able to do this characterization of flavor profile of different type of ingredient needs a lot of money. So we need funder. Funder either is coming from government agency, which realizes the huge potential based on nutrition and health of the country, or it should come from private industry, you know, private labs such as industry, they should be realizing its value and should be putting their money. It has not happened so far. So you're looking at exploring personalized nutrition and intersecting with computational gastronomy, correct? Yeah, very much. So Pranav is asking, what about the food pairing in sweets? Is it similar to food or different in regards to flavors? Human sweet meats. Yeah, desserts, basically. Desserts, yes. And I guess he's talking about the Indian context which might be heavily dairy based and, you know, have sugar. But I think actually you could maybe talk about the spices that are involved here and how different they can be. You know what? There are so many ideas that have been popped at me including, for example, dishes which are made as naivedams in temples, for example, right? The whole of all of them, can you investigate those itself, you know, as a repertoire of recipes, apart from others, tribal recipes as a set of recipes. And similar such subsets of recipes have been suggested including desserts, right? To answer your question, I have not done the analysis of this in the context of desserts, so I don't know the answer. We haven't done specifically for desserts what is the food pairing index and how different it is from other dishes. We haven't done that yet. Okay. Thank you very much. I think we've run through our time. Professor Ganesh, would you like to leave us with any last thoughts? Hopefully to do with how we can maybe apply FlavorDB in our own everyday lives from today onwards. Sure. So one is FlavorDB, another is RecipeDB. Those of you who are interested more in practical applications of computational gastronomy, you may start investigating these databases to look at what are the Flavor composition of different ingredients that go into our daily recipes. In my opinion, that will be an eye-opening exercise to begin with just to find, for example, that how sulfurous compounds are so much dominant in onion or similar ingredients which have pungent aroma that itself is a learning exercise, in my opinion. Secondly, you should be able to do food pairing by using the food pairing app which is part of FlavorDB and thereby come up with some interesting pairing by which you can change your recipes and hopefully come up with some novel tweaks in the recipes that you like already probably. And if you do so, don't forget to write to me. I have my email ID written on the database, so do that. Because if things don't work, things are not working according to your intuition or the logic, as I have proposed, don't hesitate to write back to me also. And if you have any other adventurous applications, remember this is a database that I have created from primarily from academic endeavour and has ended up becoming of applied nature. And similarly, recipe DB. So we have database of 118,000, 118,000 recipes with us, which we have broken down into its elements, ingredients, quantity, units, blah, blah. So use it, you know, find ways by which you can see, you can search recipes. Show me all recipes which have, you know, let's say peanut in it from Brazilian cuisine, which have peanut in it but doesn't have chili in it from Brazilian cuisine, or whatever cuisine that you would like to search for. You can make such queries by using recipe DB and probably you might hit upon some interesting recipes. These are real recipes. So eventually you will reach a link which can take you to original database on all recipes.com or genius kitchen.com etc. In addition to giving you insight into the recipe about what is the composition of the recipe. So those are the couple of things that I would suggest to you. And further, finally, final thoughts is this. Two things, two folks, one is of academic nature. If you think of some interesting ideas, which can be run on with the computational gastronomy domain. Feel free to write to me that this is an interesting thought or interesting experiment that can be done. Can you please work on it, like the deserts, for example, was one of the case. Can you look at food pairing index of deserts and compare it with dishes, right? You can come up with such thoughts. And finally, if you see there is an industry partner that you know of among the audiences who can be of value to me in terms of partnering. Feel free to connect me so that we can take it forward, apart from you yourself coming up with propositions of course. I can be your partner over there. We run the wood of us and we primarily work on nostalgia. One of the things that works very well for us, apart from the health aspect of the bar is that we use jaggery. And we caramelize it in butter or ghee because we clarify the butter to make ghee. And it immediately sends everybody back home to Dadi's laddus or chickpeas. So yeah, I would love to collaborate with you. This is very exciting for me and crowdsourcing flavor libraries will be fantastic I think for your research as well as for our enhanced knowledge with regards to flavors. From a nutritional aspect though, does flavor DB have the nutritional aspect matched to it? Recipe DB has it some information of nutritional value that for example, there is a recipe there are ingredients. What is the nutritional value of that particular ingredient has been estimated. There can be mistakes because this has been done using a machine learning algorithm, because we had to do it for 1,18,000 recipes right. So we did it for a few of them trained it. So there is a protocol which has gone behind it. So using that we have estimated the nutritional value. So they are available. One last thought before I end this session with you and I am has a very comprehensive nutritional database, especially with regards to Indian foods and recipes. I advise people to also take a look at that and see if there's any discrepancies. Thank you very much Professor Ganesh. What a pleasure having you here. I really enjoyed your talk and I'm really looking forward to gastronomica. And I'm really looking forward to hopefully yelling out to Alexa that you know this is what I have in my fridge tell me what I can make with the best food pairing index possible. It's my pleasure to thank you. Thanks a lot.