Will A.I. Unlock The Secret To The Best Diet For Human Well-Being?
Photo by Dennis Klein on Unsplash
If you’re even remotely interested in diet, you must have read about tons of diet programs that promise the same things — weight loss, muscle building, an active mind, lower stress, and so on. Some believe a low-fat diet works, while others believe a high-fat-low-carb diet is the best.
The opinions of staunch proponents of the myriad diets don’t really tell us a lot about the effectiveness of a diet. Eran Segal, a computational biologist believes that if diet A is better than diet B then enough people should show that definitely — no opinions or beliefs. Just hard facts.
Alas, no such diet has been found yet, even after decades of research on the topic. According to Segal, we still struggle to answer the basic question — “What’s the best diet for humans?”
But he says that’s not because we’re dumb, but because we’re asking the wrong question. What if there’s no single best diet for all of us? What if our genetic makeup, gut bacteria, lifestyle, etc causes some diets to be better for us than others? What if some diets that are considered “bad” outrightly end up working for a specific group of people?
This ideology has largely been ignored due to our focus on the types of food rather than the person eating it. And this is exactly what Eran focused on with his colleague Eran Elinav and several graduate students from Weizmann Institute of Science.
The Metric of Health
Before even beginning their analysis, Segal and the team searched for the best metric they can use to measure if a particular diet is better for participants than the others.
Many studies examine weight loss, risk of heart disease, or other similar metrics. The problem with these metrics, however, is they’re affected by many other factors unrelated to diet and take many weeks to show changes.
Segal on the other hand chose a metric that is relevant and easy-to-measure — changes in blood glucose levels after a meal. High blood glucose levels after a meal promote weight gain and hunger by converting excess sugar into fat. Moreover, this fast flow of glucose in the blood often causes the body to release too much insulin which can lower our glucose levels to a point where we start to feel hungry again.
And since the average person eats around 50 meals a week, it allows the team to measure glucose responses to 50 meals in just a single week.
The Experiment Begins
After gathering about 1,000 people, they started tracking the glucose levels of people 50 times a week. During that week, participants logged everything they ate on a mobile app. Having collected data for over 50,000 different meals, this study became the largest one conducted on this problem.
So what did they find after analyzing the data? Well, needless to say, they saw trends. For instance, more carbohydrates increase the meal glucose response.
But more surprisingly, for every trend they found, there were many people who differed from it. Meaning, when the same person ate the same meal on different days, the response was very similar. But when different people ate the same meal, the response was very different.
For instance, they found, white bread induced almost no effect on the blood sugar levels of most people, but in others, it induced huge spikes.
This was true for every food they tested — pizza, rice, sushi, you name it.
Just as the team hypothesized, the response depended not as much on what food one eats, but on who was eating it.
The results of such a large dataset convinced Segal and his team that diets that maintain normal blood glucose levels must therefore be personally tailored to the individual.
Moreover, it strengthened their belief about the flawed nature of the current nutritional paradigm that is in search of the best diet for everyone.
Personalized Dietary Recommendations At Scale
If the best diet differs for every person, how can we decide what to eat and what to skip? Fortunately, machine learning can guide us.
After collecting data on blood glucose levels and the kind of food the participants ate, the team also collected information about the people themselves.
They started looking for parameters that could perhaps explain why there was such variability in the meal glucose responses of all these people. These parameters include weight, age, height, physical activity, medical background, DNA sequencing, gut bacteria, and so on.
They used advanced machine learning algorithms on this clinical data set to automatically search for rules that predict personalized glucose responses to meals. For example, one such rule can be that if you’re over 60 and you inhibit certain bacterial species, then your response to ice cream will be high.
The algorithm thus created tens of thousands of such rules from the data. Soon, this algorithm could take any person, even people outside the participants, and predict the response to arbitrary meals with high accuracy.
As a final test of this algorithm, the team bought in new participants and asked the algorithm to predict two diets for each person — one bad and one good. Naturally, the bad diet had foods with high meal glucose responses and vice-versa.
One key point to be noted is that all diets were completely identical in the number of calories a person was eating.
The results for this experiment were also positive. The good diet led to low glucose responses even though it included foods like ice cream that are considered “bad” by most people. This week also induced beneficial effects in people that may persist way beyond the experiment, which was great.
The Take-Home Message
The final message, after all this research, is that your diet may fail or succeed simply because it did not take information about you as an individual into account. What Segal says we can do with this information is measure your personal glucose responses to certain meals using simple glucose devices available in common shops.
Since many people may not go to such lengths, Segal and the team are working to make these algorithms available to everyone so you can send a sample of your microbiome along with clinical data and receive personalized dietary advice.
So will this be the new way humans decide what to eat and what to discard? Would you? I’d love to know what you think!