The lamb leg wants to kill me. The stress it will cause to my blood fats and my gut microbiome does not make up for its neutral effect on my glucose levels. I’m hungry, though, and it smells delicious. What should I do?
I’m trialing the latest app in a burgeoning industry of personalized-nutrition products, which are based on cutting-edge science clinically shown to identify precisely the foods that are good for us to eat and when to eat them. This app, Zoe, comes from a highly regarded scientific team at the University College London. Among many other accomplishments, they have for years conducted one of the world’s largest twins studies, uncovering many key mysteries about the relationship between genes and environment, including the foods we eat.
In my case, Zoe tells me the lamb scores a mere 15 points out of 100. These points, generated by a computer algorithm customized to my own metabolism, summarize how good or bad something is for me to eat at this moment. My goal is to maintain a daily average of 75, and so far I’m slightly above that, so I don’t think I can take the hit.
Normally, I would have two choices: Switch yet again to a spinach salad with no dressing, or forget the dietary advice and worry about the consequences later. Unfortunately for many of us, later has already happened. Our little dietary sins, compounded daily, end up costing far more than we think, in obesity, poor sleep, low energy, skin and digestive problems, and an untold number of other health issues, many of which would go away if we could just eat better.
Zoe gives me a third choice: because its scoring system is based on a complex machine-learning model trained on more than 1,000 subjects eating millions of meals, it can recommend food combinations, some of which can make up for what would be unhealthy if eaten on its own.
So I add some black beans (+35), wilted spinach (+10), and a honeycrisp apple (+12) to my meal, I’m magically I’m back to my goal for the day. I can have my lamb and eat it too!
Zoe is the latest in a series of products, including Viome and DayTwo (which I wrote about previously in NEO.LIFE), that offer direct-to-consumer access to newly discovered insights about the human microbiome, the collection of tiny microbial organisms inside you that science is just beginning to explore. In fact, maybe only 1% of the genes in your body come from your parents; the rest are microbial cells, trillions of them, all invisibly tiny, but collectively enough to weigh about five pounds—roughly as much as your brain. The thousands of unique bacterial species in your gut play a major role in your digestive health, and have been linked to nearly every chronic disease, from Alzheimer’s and cancer to diabetes and allergies. The vast majority of these microbes are benign, though, and far more of them are beneficial than pathogenic. Without your bacterial benefactors, your body wouldn’t produce enough of key metabolites, including vitamin K or the sleep-inducing melatonin.
The plummeting cost of gene sequencing has brought consumer prices to microbiome technology, letting anyone peer into their microbial health for as little as $100 from companies like Thryve or Psomagen. More comprehensive products from Viome ($349), DayTwo ($499), and now Zoe ($379) offer clinical-grade diagnostics for people who want a state-of-the-art look at their microbes.
I’ve tried each of these products, and Zoe is the most thorough. The process starts with them sending you a boxful of diagnostic tests I did at home. Besides the gut microbiome test, which involves harvesting a tiny bit of stool, they sent me a continuous glucose monitor (CGM, like the one I wrote about previously for NEO.LIFE), a box of specially prepared muffins, and a handful of dried spot blood (DSB) cards to prick my finger at various pre-determined times. The muffins contain carefully selected ingredients so Zoe’s lab can tell how my body responds to various levels of sugar and fat, as measured by the CGM and the DSB cards.
The downside to all these tests is that the initial diagnostic process takes longer and is more cumbersome than the comparatively simple “poop-in-a-box” microbiome-only tests. In the pre-release version I tried, I had to collect about 10 days’ worth of data—eating muffins, pricking my fingers—before I could submit my tests, and then it’s another month or so for the results. But the final insights are the most full-featured I’ve seen yet, with personalized dietary advice distilled into in an app that is exceptionally easy and useful.
I think of myself as a reasonably healthy omnivore: I’m at about my ideal weight and exercise regularly, with no serious medical conditions. Although I’m no nutritionist, my previous experiments in observing my microbiome, CGM, and other food tracking probably makes me more aware of what I eat than the average person. Still, Zoe taught me a lot. Here are some of my key takeaways:
- Calories are irrelevant. Zoe doesn’t even bother counting them, except as a way to track nutrient density. The app rates foods by their nutrients per 100 kcal, ignoring the total.
- Vegetables, vegetables, vegetables! To get to my 75+ score, I need to eat at least five unique fruits or vegetables per day.
- Processed food is never okay. Even those expensive branded protein and fiber bars all score poorly for me.
- Timing makes a difference, especially when it comes to fats. The app uses the insights gleaned from those muffins and DSB tests to help me with meal timing. In my case, the tests show I’m a “poor” blood fat metabolizer who should schedule any fat-heavy meals about six hours apart. My daily personalized score will suffer if I eat sooner.
Finally, food combinations matter, as I learned with the lamb example. Studies of how various foods affect the microbiome often take place in isolation: Half the study participants eat X, half don’t, and we compare how their microbiomes differ. Although this makes a complicated process much simpler, it ignores the reality that people don’t eat X—they eat meals. Who would be satisfied only eating carrots for dinner, or plain pasta, or, for that matter, lamb? Some of the ingredients I learned to eat together included apples and almond butter, and I learned to avoiding mixing bananas and coconut milk.
To work out which food combinations work best together, the scientists behind Zoe tracked everything their study participants ate, and then ran complex machine-learning algorithms to piece out which combinations are associated with each change in the microbiome. After you return your testing kit, their lab uses the same algorithm to generate a diet suggestion that’s too complicated to simplify into a single description like “low-carb” or “paleo” or even “Whole30.” Only one person in the world fits this diet, and it’s you.
The diet Zoe recommends is highly personalized. Just about everyone can benefit from more fruits and vegetables, but you may learn some surprises about which ones are better or worse for you. The app got me to look at beans more seriously, for example, which are discouraged on popular paleo or ketogenic diets, but highly recommended in all of the microbiome-based tests I did from Viome, DayTwo, and now Zoe. As much as I enjoy rice, I can’t deny that all my tests forbid it, even the “whole grain” varieties.
Food tracking is the Achilles heel of any personalized nutrition app, not just because it’s a pain to enter all your meals, but because often the healthiest choices are those you make from scratch, at home, with multiple ingredients; you can’t scan the barcode for a slice of your legendary lasagna, and painstakingly going through and logging every ingredient in a complicated recipe can be cumbersome.
Zoe does as good a job as possible to alleviate this. The barcode reader works great for packaged foods, and the overall database is extensive and usually had an entry for anything I tried. Some important foods were missing in the beta version (coffee, wine, garlic, and other spices), but most ethnic foods are there (such as kimchi, kvass, and dhal).
Because the spacing of your meals is important, Zoe requires that you timestamp everything you eat. For simple meals, it’s not hard to enter your food as you eat it, but if you’re too busy to use the app immediately, or if you’re in a situation where pulling out your phone would be an awkward interruption, you’ll need to remember both the contents and timing of your meal later. It would be nice if the app had an option to include a quick timestamped photo of my meal to make this easier.
Our little dietary sins, compounded daily, end up costing far more than we think.
Still, Zoe handles this as well as could be expected: The app groups each food into “breakfast,” “lunch,” “dinner,” and “snack,” automatically adding the current time to every entry. Because of my poor fat metabolism, my daily score is penalized if I eat too much fat too quickly, and the app helpfully tells me what time I’ll be able to eat a ball of burrata again.
One consequence of Zoe’s all-encompassing but sometimes opaque personalized model is that what you enter can often unexpectedly cause your results to swing wildly. Is this because certain combinations of food really do dramatically affect me? Or is it just a bug in the software? It can be hard to tell, and some days I wished I could open up the black box of the algorithm to check.
For example, it doesn’t take a fancy app to tell me that broccoli is good for me (99). And I get why eating it with cheese (59) drops my score to 80. But why, when I throw in an apple (89), do I shoot all the way back up to 92? The final number at the end of the day depends on these food combinations plus the variety and timing. It’s nice that I can “simulate” the various alternatives beforehand, but I wish the app would give me more hints of what to try.
So after faithfully following Zoe’s advice for four weeks, what’s the bottom line? Like I said, I’m already pretty healthy, so I can’t point to the kind of massive improvement that other users might experience (although my before/after blood tests showed my cholesterol dropped by about 10%). Still, I’m grateful that the scientists behind Zoe have chosen to make their insights available at consumer prices to anyone who wants to try it.
Ultimately the biggest eye-opener for me was a new appreciation for food combinations. We don’t eat ingredients—we eat food, and the Zoe algorithm understands that. The startup process was cumbersome, with lots of tests to take and some time-consuming data entry about my meals, but after a month of daily usage, it’s becoming second nature. Lamb chops no longer scare me.