The algorithm to a perfect diet- AI has answers

Despite decades of research, government recommendations and temporary fixations with the ‘best diets’ ever, we do not know much about nutrition.  There are meal plans out there that have recommendations for people with specific restrictions and needs such as for diabetes , which includes eating 3 meals a day, eating foods with high fiber, lowering sugary foods and so on.
These meal plans, however, are based on empirical consumer studies, rather than individualized requirements.

This is a two blog series, the first introduces personalized nutrition and how it works. The second talks in detail about challenges and opportunities for precision nutrition. Further, I shall talk about letting Artificial Intelligence (AI_ decide what you eat, about how machine learning and deep learning are redefining nutrition. And perhaps, in the future, you would be able to take a simple blood test and come up with a fools proof diet that would actually work.

Personalized nutrition: What is it?

Simply put, ‘personalized nutrition’ is the dietary analog of personalized medicine. It is a way of customizing nutrition based on assessing an individual’s genetics, dietary habits, physical exercises, gut microbiome, and sleep behaviours (1).

Current diets and everything that is wrong with them.

Paleo, Mediterranean, Atkins, keto and so on. These are some very common diet buzz words thrown around by men and women who are trying to get ‘fit’. These diets have themes such as eating foods available for consumption to humans in the Palaeolithic era, eating foods commonly consumed in Greece, Southern Italy, and Spain in the 40s and 50s, cutting carbs and rewiring your body to ketosis, to name a few. Furthermore, these have another common theme-– the general assumption that it would work for most common people.

Most of these studies are highly observational. They depend on people reporting from memory, through questionnaires or by maintaining a diary — of how much of what they consumed. This has a basic flaw, people, like their memories, cannot be trusted to provide an accurate and complete picture. These provide a correlation, not a cause-effect relationship, and moreover, they often contradict each other!

Why you can’t paint it all with the same brush?

Let me give you more reasons why these generalized assumptions fail us.
Post-meal glucose levels are being increasingly being recognized as risk factors for cardiovascular disease and mortality. In an interesting paper, smartly entitled ‘Siri, what should I eat?‘ Schwatzenberg and Turnbaugh explain that the postprandial plasma glucose levels vary vastly among individuals. Some factors affecting this variability are a combination of micro, macro-nutrients and numerous other variables such as anthropometrics, meal times, sleep/wake cycles, physical activity, lifestyle, insulin sensitivity/ resistance, and the gut microbiome (2).

This simple metric itself translates to a monumental interindividual variability. I haven’t even added the gut microbiome composition, ethnicity, sex, heredity, age, choice, quality, and quantity of food, and most importantly the genes to this mix!

MIND BOGGLING!

So I think I can safely assume that I have established that we shouldn’t generalize and there is a pressing need to specifically design food, like medicine.

Quest for answers

Zeevi et. Al., 2015 designed a framework to methodically address the above case. The authors collected data from 800 people about- the time of each meal, food and beverage consumed, physical activity, sleep/ wake cycles, and anthropometric measures using a mobile application. The patients also had their blood and gut microbiome assessed, and blood glucose levels studied for a week. They ate standardized meals provided by the researchers with nearly 47000 kinds of meals . In total, they measured about 1.5 million data points. That’s a lot!

They used machine learning, a kind of AI to understand this data set and to figure out what led to these differences in each individual. These results were validated with an independent 100 person cohort. Further, they conducted a blinded, randomized, controlled dietary intervention based on this algorithm. They reported significantly lower postprandial glucose levels and alterations in the gut microbiota!

The authors have a commercially available product that uses a similar algorithm to personalize your diet!

In another such program, Adults from 7 European countries were recruited to an Internet-delivered intervention called Food4Me. They randomized multiple variables and outcomes-based on dietary intake, anthropometry, and blood biomarkers were measured for 3-6 months. They concluded that internet-delivered personalized nutrition intervention produced larger and more appropriate changes in dietary behavior against conventional diets (4). Such proofs of concept pave way for a larger pilot scale investments.

What’s happening now..

Nestle is pioneering in this field. They are piloting a personalized nutrition program in Japan called the ‘ Wellness Ambassador’. This program combines artificial intelligence, DNA testing and meal analysis to collect information on consumers’ diet and health. The study involves 100,000 people shareing pictures of their meals using a chat application that uses AI and recommends dietary, activity changes, and personalized supplements.

For now, here’s what siri suggests I do, I am looking forward to some personalized advice soon!

That’s it for this blog folks. I hope you’re as excited about this as I am . Personalized nutrition still has a long way to go, but is already showing promising results. I shall end this blog here and share more about some opportunities and challenges precision nutrition must face before being accessible by all.

This is Roo signing off, thank you.

Here are some interesting resources:

1. Kussmann, M., & Fay, L. B. (2008). Nutrigenomics and personalized nutrition: science and concept.
2. von Schwartzenberg, R. J., & Turnbaugh, P. J. (2015). Siri, What Should I Eat?. Cell, 163(5), 1051-1052.
3. Zeevi, D., Korem, T., Zmora, N., Israeli, D., Rothschild, D., Weinberger, A., … & Suez, J. (2015). Personalized nutrition by prediction of glycemic responses. Cell, 163(5), 1079-1094.
4. Celis- Morales, C., Livingstone, K. M., Marsaux, C. F., Macready, A. L., Fallaize, R., O’Donovan, C. B., … & San-Cristobal, R. (2016). Effect of personalized nutrition on health-related behavior change: evidence from the Food4me European randomized controlled trial. International journal of epidemiology, 46(2), 578-588.