Computational Gastronomy
GANESH BAGLER
http://cosylab.iiitd.edu.in
An emerging data science of food, flavors and health
An aspiring astronomer
Dieter Spears | Inhaus Creative
From astronomy to gastronomy
IHBT
SFI
CCMB
Computational Gastronomy is a data science
that blends food, data, and computation for
data-driven food innovations
‘Catching Fire—How cooking made us human’ by Richard Wrangham
Image Credits: Chloe Cushman
Image Credits: Chloe Cushman
Innovation [noun]
make changes in something established,
especially by introducing new methods, ideas, or products.
The Enquirer, Vraja Kishor's Blog
DATA
Making Food Computable
Why do we eat what we eat?
Why do we combine ingredients
in our recipes the way we do?
Food Pairing Hypothesis
Ingredients that taste similar tend to be used together
in traditional recipes
Ahn et. al, “Flavor network and the principles of food pairing”, Scientific Reports (2011).
A Jain, NK Rakhi, G Bagler*, “Spices form the basis of food pairing in Indian cuisine”, arXiv:1502.03815 (2015).
Ganesh Bagler*, ‘The digital way forward for the Indian Cuisine’ (2019). Image Courtesy: FSSAI EatRight Mela Poster, Dec 2018
Recipes & Ingredients
2543 Traditional Indian Recipes (TarlaDalal)
Regional cuisines: Bengali, Gujarati, Jain, Maharashtrian,
Mughlai, Punjabi, Rajasthani, South Indian.
A Jain, NK Rakhi & G Bagler,* “Analysis of Food Pairing in Regional Cuisines of India”, PLoS ONE, 10(10): e0139539(2015).
Olfactory
fla•vor = smell + taste
Gustatory
Fenaroli’s Handbook
Ahn et. al, Sci.Rep. (2011)
Uniform Food Pairing
Western Cuisines
𝑅𝑎𝑛𝑑𝑜𝑚 𝐶𝑢𝑖𝑠𝑖𝑛𝑒
Indian Cuisine
Contrasting Food Pairing
Jain et. al, PLoS ONE (2015)
Spices are key to the food pairing in Indian cuisine
Food Pairing
12
10
Spice
Vegetable
Fruit
8
Plant derivative
Nut/seed
Cereal/Crop
6 Dairy
Plant
Herb
4 Original recipes
A Jain, NK Rakhi, G Bagler*, “Spices form the basis of food pairing in Indian cuisine”, arXiv:1502.03815 (2015).
Spice
The Taste of India
Anupam Jain, NK Rakhi and Ganesh Bagler*, arXiv (2015); Anupam Jain, NK Rakhi and Ganesh Bagler*, PLoS ONE (2015).
A Jain, NK Rakhi and G Bagler*, arXiv (2015); A Jain, NK Rakhi and G Bagler*, PLoS ONE (2015).
Best of 2015
MIT Technology Review
A Jain, NK Rakhi and G Bagler*, arXiv (2015); A Jain, NK Rakhi and G Bagler*, PLoS ONE (2015).
Making Food Computable
————————————————
Data-Driven Food Innovations
RecipeDB
Batra et al., ‘RecipeDB: A resource for exploring recipes’, Database (2020).
Navjot Singh and Ganesh Bagler*, IEEE International Conference on Data Engineering, DÉCOR, Paris (2018).
The Tree of World Cuisines
G Bagler*, N Singh, R Tuwani, et. al. (2019). Jain and Bagler, Physica A (2019).
R Tuwani, N Sahu, G Bagler*, IEEE International Conference on Data Engineering, DÉCOR, Macau, China (2019).
T Sharma, et al., ‘Hierarchical Clustering of World Cuisines’, International Conference on Data Engineering (2020).
30
10
Credits: Wiki; Wired
Y Agarwal, D Batra and G Bagler*, ‘Building Hierarchically Disentangled Language Models for Text Generation with Named Entities’, COLING (2020).
https://cosylab.iiitd.edu.in/ratatouille
https://cosylab.iiitd.edu.in/ratatouille
http://cosylab.iiitd.edu.in/flavordb
http://cosylab.iiitd.edu.in/flavordb/
Nucleic Acids Research (2017). Complex Systems Laboratory, IIIT-Delhi
Taste Prediction
Bitter Sweet
Scientific Reports (2019)
https://cosylab.iiitd.edu.in/bittersweet
How sweet is that?
Preliminary results
A Sharma, S Jain, R Tuwani and G Bagler*, ‘Machine learning models for predicting sweetness of a compound,’ manuscript under preparation (2020).
Computational Food
The
TheDrug-Maker’s
Food Maker’sGuide to the
Guide to Galaxy
the Galaxy
How
HowMachine
MachineLearning
Learningand
andBig
BigData
Datacan
arehelp chemists
helping search
chemists
search
the vastthe vast chemical
chemical universeuniverse forfood
for better better medicines
Adapted from the ‘The Drug Maker’s Guide to the Galaxy,’ A. Mullard, Nature (2017)
Why spices? Broad-spectrum benevolence of culinary herbs and spices
“Data-driven analysis of biomedical literature suggests broad-spectrum benefits of culinary herbs and spices,”
Rakhi NK, Tuwani R, Mukherjee J, Bagler G, PLoS ONE 13(5): e0198030 (2018).
R Tuwani, Rakhi NK, N Garg and G Bagler*, under review; R Tuwani, Rakhi NK, and G Bagler*, under review.
Personalized Nutrition
Gut Microbes
Blood tests
Questionnaires
Body Mass Index
Food diary
Design personalized diet to
lower glucose levels
ED Sonnenburg and JL Sonnenburg, Nature, 528, 484 (2015).
Zeevi et al., “Personalized nutrition by prediction of glycemic response”, Cell, 163, 1079 (2015).
Computational Gastronomy — Data-driven food innovations
Computational Gastronomy = Food + Artificial Intelligence
“ The discovery of a new dish confers
more happiness on humanity, than the
”
discovery of a new star.
Jean Anthelme Brillat-Savarin
Computational Gastronomy is a data science
that blends food, data, and computation for
data-driven food innovations.
DATA
Making Food Computable
Four major aspects of Computational Gastronomy Data
1. Recipes – Information of recipes that reflects its
constituent ingredients.
Further, this could also include details of the temporal
sequence of ingredients, the quantity of each ingredient,
their state (chopped, ground, etc.), size and the method
of processing ingredients (cook, boil, fry, saute, etc.).
Four major aspects of Computational Gastronomy Data
2. Flavor Profiles of Ingredients: Every natural
ingredient that is used in recipes is primarily selected
based on its flavors—taste and odor.
The flavor profile of the ingredient represents the set
of empirically reported flavor molecules.
Four major aspects of Computational Gastronomy Data
3. Health Associations of Ingredients:
Each ingredient may be associated with the health
impacts as identified from experimental studies
linking them to diseases.
Four major aspects of Computational Gastronomy Data
4. Nutritional Profile of Ingredients:
Each ingredient can be associated with its
nutritional profile, including macro- as well as
micro-nutrients.
Potential Data-Driven Food Innovations
Novel Recipes Design: Cooking is a combinatorial
system that combines the raw material (ingredients)
and ways of processing them (cooking rules). Given
that the average recipe uses 10 ingredients and that
there are around 1000 unique ingredients, one could
create around 1030 recipes simply with the power of
combinations. Not every such ‘recipe’ would be
palatable, leave alone delicious. Knowing the patterns
in the traditional recipes from various cuisines, one
could filter recipes that conform to the cultural norms.
This requires the use of statistics, data/patterns mining,
and natural language processing.
Potential Data-Driven Food Innovations
Culinary Fingerprints: Starting with the tripartite
data of recipes, their constituent ingredients and
further the flavor profiles of ingredients, one could
mine for patterns in traditional recipes to obtain the
‘culinary fingerprints’ of cuisines. These fingerprints
represent the characteristic pattern of ingredient co-
use as well as the idiosyncratic flavor patterns of a
cuisine depicting the culinary preferences of a
population. Such insights are of value for food/recipe
design for targeted audiences.
Potential Data-Driven Food Innovations
Taste/Odor Prediction: Large compilation of flavor
molecules along with their taste/odor attributes
presents an interesting case for machine learning-
based taste and odor prediction algorithms. Here
the battery of molecular properties (starting from
simple ones such as number of atoms, molecular
weight to all the way up to molecular fingerprints
comprising of thousands of attributes) are features
which could be used for building statistical models
for classification into taste and odor classes.
Potential Data-Driven Food Innovations
De novo design of flavor/taste molecules: Along
with in silico methods for generation of organic
compounds, taste/odor classification models can
enable de novo synthesis of compounds of
desirable profile.
Potential Data-Driven Food Innovations
Healthy and Sustainable Diet Design:
Systematically curated data of health impacts of
ingredients, along with rules of making recipes, can
be used for designing potentially healthy recipes.
Interestingly, such strategies could be used to
propel the use of ingredients for promoting a
sustainable food culture.
Impact of Technology on Food
Scientific American, June 2015
Ganesh Bagler
Center for Computational Biology, IIIT-Delhi, New Delhi.
Ingredient Categories: Forms and Shapes
• Vegetable: All kind of leafy vegetables (spinach,
fenugreek), beans (kidney bean, red beans), tubers
(potato, turnip, beetroot) and such.
• Spice: Flavorful ingredients primarily from root, bark
and seeds. Examples: cumin, clove, cardamom,
cinnamon, turmeric etc.
• Herb: Flavorful ingredients primarily from leaves and
stems of plants. Examples: coriander, thyme, rosemary,
mint, basil etc.
• Pulse: The fruit or seed. Examples: chickpeas, green
gram, back gram, pigeon pea etc.
Ingredient Categories: Forms and Shapes
• Cereal/Crop: Examples: rice, oat, sorghum, barley etc.
• Plant: Examples: tea, thistle, pine, hop
• Fruit: Examples: apple, orange, guava, grape, papaya
etc.
• Flower: rose, jasmine, lavender, artichoke etc.
• Nut/Seed: Examples: peanut, walnut, hazelnut, filbert
etc.
• Dairy: Example: milk, curd/yogurt, butter, buttermilk,
cream, cheese, whey etc.
Ingredient Categories: Forms and Shapes
• Meat: Examples: beef, pork, mutton, chicken etc.
• Fish: Examples: salmon, codfish, tuna, catfish, bluefish
etc.
• Seafood: Examples: lobster, mollusk, oyster, prawn,
shrimp, kombu, kelp etc.
• Beverage-Alcoholic: Examples: rum, gin, wine, whisky
etc.
• Beverage: fruit juice, soft drink, milk shake etc.
• Additives: salt, sugar, Monosoldium glutamate (MSG)
Scientific American, June 2015
Scientific American, June 2015
Scientific American, June 2015
Scientific American, June 2015
Processed Food
It is the dark force, we’re told, behind the obesity
epidemic, and the death of the family farm.
But humans have been “processing” food ever
since we learned how to cook, preserve, ferment,
freeze, dry or extract.
Processed food has powered the evolution of the
species, the expansion of empires, the
exploration of space.
Scientific American, June 2015
Noodles
Pickles 2000BC
2400BC
Palm Oil
3000BC
Roasted Meat
1.8mya
Bread Olive Oil
Cheese 4500BC
12000ya Tortillas 5000BC
6700BC
Beer Wine
7000-3000BC 5400BC
Scientific American, June 2015
Peanut Butter
15th Century
Chocolate
1900BC
Salt Cod
10th Century
Bacon Tofu
1500BC 965AD
Sushi
700AD
Jiang
1000BC
Sugar Mustard Kimchi
500BC 400AD 700AD
Scientific American, June 2015
Carbonated Water
1767
Corn Flakes
Coffee 1894
15th Century
Monosodium Glutamate Lab Grown Meat
1808 2013
High Fructose
Spam Chicken Nuggets Corn Syrup Tang
Plumpy’Nut
1926 1950 1957 1959
1996
Scientific American, June 2015
Google Images
The
Impossible
Burger
Cooked – Michael Pollan
The Omnivore’s Dilemma –
Michael Pollan
The Botony of Desire– Michael
Pollan
Fast Food Nation – Eric
Schlossler
Recipe Data Analytics
Recipes: Data and Analysis
• What is a cuisine??
– Geo-cultural defn; Geographical indication; the Rosogulla conundrum
– History of recipe documentation; Missing/lost recipes
• Coarse-grained data & spherical cows
• Recipe Structure
– Ingredients & Cooking Instructions
• Recipe Size Distribution
– The normal distribution & inferences
• Ingredient Popularity
– The generic pattern across cuisines
• Category Composition
• Mining Patterns in Recipes
– Ingredient pairing
– Frequent Itemset Mining
Recipes Data Structure
_________________________________ RECIPE _________________________________
INGREDIENTS SECTION
100gsms, potato, sliced
50gms, capsicum, finely chopped
50 gms Onions, chopped
100ml groundnut oil
10gms, cumin, crushed
10gms, turmeric
20gms, chilly power
To taste, salt
COOKING INSTRUCTIONS
1. Add oil to pan and heat it for 3-4 minutes.
2. Add cumin and then add onion. Fry for 3-4 minutes
3. Add potato and capsicum. Fry for 8-10 minutes.
4. Add turmeric, cumin and salt. Mix thoroughly.
5. Keep the mixture on low heat for another 5 minutes.
Recipes & Ingredients
2543 Traditional Indian Recipes (TarlaDalal)
Regional cuisines: Bengali, Gujarati, Jain, Maharashtrian,
Mughlai, Punjabi, Rajasthani, South Indian.
A Jain, NK Rakhi & G Bagler,* “Analysis of Food Pairing in Regional Cuisines of India”, PLoS ONE, 10(10): e0139539(2015).
Recipes Data
Recipes Data Structure
Recipe-01 : Ingredient-01, Ingredient-02, Ingredient-15, Ingredient-19, Ingredient-06
Recipe-02 : Ingredient-06, Ingredient-12, Ingredient-08, Ingredient-15
.
.
.
.
.
.
.
.
.
.
.
.
Recipe-100: Ingredient-06, Ingredient-11, Ingredient-15, Ingredient-13
Recipes Data Structure
Recipe-01 — Ingredient-01
Recipe-01 — Ingredient-02
Recipe-01 — Ingredient-15
Recipe-01 — Ingredient-19
Recipe-01 — Ingredient-06
Recipe-02 — Ingredient-06
Recipe-02 — Ingredient-12
Recipe-02 — Ingredient-08
Recipe-02 — Ingredient-15
.
.
.
.
.
.
Recipe-100 — Ingredient-06
Recipe-100 — Ingredient-11
Recipe-100 — Ingredient-15
Recipe-100 — Ingredient-13
Recipes Data Statistics
The recipe size distribution of recipes is observed to follow a
normal distribution.
Recipes Data Statistics
The nature of the Frequency-Rank statistics is observed to consistently present
with a power law across the world cuisines.
Recipes Data Statistics
Recipes Data Statistics
