Which foods are actually healthy for your individual gut microbiome? Several companies are offering personalized dietary guidance based on your test results, but their answers in one experiment turned up with some conflicting advice.
I recently got on the scale to weigh myself, thinking I've got to eat better. With so many trendy diets today claiming to improve health, from Keto to Paleo to Whole30, it can be confusing to figure out what we should and shouldn't eat for optimal nutrition.
A number of companies are now selling the concept of "personalized" nutrition based on the genetic makeup of your individual gut bugs.
My next thought was: I've got to lose a few pounds.
Consider a weird factoid: In addition to my fat, skin, bone and muscle, I'm carrying around two or three pounds of straight-up bacteria. Like you, I am the host to trillions of micro-organisms that live in my gut and are collectively known as my microbiome. An explosion of research has occurred in the last decade to try to understand exactly how these microbial populations, which are unique to each of us, may influence our overall health and potentially even our brains and behavior.
Lots of mysteries still remain, but it is established that these "bugs" are crucial to keeping our body running smoothly, performing functions like stimulating the immune system, synthesizing important vitamins, and aiding digestion. The field of microbiome science is evolving rapidly, and a number of companies are now selling the concept of "personalized" nutrition based on the genetic makeup of your individual gut bugs. The two leading players are Viome and DayTwo, but the landscape includes the newly launched startup Onegevity Health and others like Thryve, which offers customized probiotic supplements in addition to dietary recommendations.
The idea has immediate appeal – if science could tell you exactly what to make for lunch and what to avoid, you could forget about the fad diets and go with your own bespoke food pyramid. Wondering if the promise might be too good to be true, I decided to perform my own experiment.
Last fall, I sent the identical fecal sample to both Viome (I paid $425, but the price has since dropped to $299) and DayTwo ($349). A couple of months later, both reports finally arrived, and I eagerly opened each app to compare their recommendations.
First, I examined my results from Viome, which was founded in 2016 in Cupertino, Calif., and declares without irony on its website that "conflicting food advice is now obsolete."
I learned I have "average" metabolic fitness and "average" inflammatory activity in my gut, which are scores that the company defines based on a proprietary algorithm. But I have "low" microbial richness, with only 62 active species of bacteria identified in my sample, compared with the mean of 157 in their test population. I also received a list of the specific species in my gut, with names like Lactococcus and Romboutsia.
But none of it meant anything to me without actionable food advice, so I clicked through to the Recommendations page and found a list of My Superfoods (cranberry, garlic, kale, salmon, turmeric, watermelon, and bone broth) and My Foods to Avoid (chickpeas, kombucha, lentils, and rice noodles). There was also a searchable database of many foods that had been categorized for me, like "bell pepper; minimize" and "beef; enjoy."
"I just don't think sufficient data is yet available to make reliable personalized dietary recommendations based on one's microbiome."
Next, I looked at my results from DayTwo, which was founded in 2015 from research out of the Weizmann Institute of Science in Israel, and whose pitch to consumers is, "Blood sugar made easy. The algorithm diet personalized to you."
This app had some notable differences. There was no result about my metabolic fitness, microbial richness, or list of the species in my sample. There was also no list of superfoods or foods to avoid. Instead, the app encouraged me to build a meal by searching for foods in their database and combining them in beneficial ways for my blood sugar. Two slices of whole wheat bread received a score of 2.7 out of 10 ("Avoid"), but if combined with one cup of large curd cottage cheese, the score improved to 6.8 ("Limit"), and if I added two hard-boiled eggs, the score went up to 7.5 ("Good").
Perusing my list of foods with "Excellent" scores, I noticed some troubling conflicts with the other app. Lentils, which had been a no-no according to Viome, received high marks from DayTwo. Ditto for Kombucha. My purported superfood of cranberry received low marks. Almonds got an almost perfect score (9.7) while Viome told me to minimize them. I found similarly contradictory advice for foods I regularly eat, including navel oranges, peanuts, pork, and beets.
Contradictory dietary guidance that Kira Peikoff received from Viome (left) and DayTwo from an identical sample.
To be sure, there was some overlap. Both apps agreed on rice noodles (bad), chickpeas (bad), honey (bad), carrots (good), and avocado (good), among other foods.
But still, I was left scratching my head. Which set of recommendations should I trust, if either? And what did my results mean for the accuracy of this nascent field?
I called a couple of experts to find out.
"I have worked on the microbiome and nutrition for the last 20 years and I would be absolutely incapable of finding you evidence in the scientific literature that lentils have a detrimental effect based on the microbiome," said Dr. Jens Walter, an Associate Professor and chair for Nutrition, Microbes, and Gastrointestinal Health at the University of Alberta. "I just don't think sufficient data is yet available to make reliable personalized dietary recommendations based on one's microbiome. And even if they would have proprietary algorithms, at least one of them is not doing it right."
There is definite potential for personalized nutrition based on the microbiome, he said, but first, predictive models must be built and standardized, then linked to clinical endpoints, and tested in a large sample of healthy volunteers in order to enable extrapolations for the general population.
"It is mindboggling what you would need to do to make this work," he observed. "There are probably hundreds of relevant dietary compounds, then the microbiome has at least a hundred relevant species with a hundred or more relevant genes each, then you'd have to put all this together with relevant clinical outcomes. And there's a hundred-fold variation in that information between individuals."
However, Walter did acknowledge that the companies might be basing their algorithms on proprietary data that could potentially connect all the dots. I reached out to them to find out.
Amir Golan, the Chief Commercial Officer of DayTwo, told me, "It's important to emphasize this is a prediction, as the microbiome field is in a very early stage of research." But he added, "I believe we are the only company that has very solid science published in top journals and we can bring very actionable evidence and benefit to our uses."
He was referring to pioneering work out of the Weizmann Institute that was published in 2015 in the journal Cell, which logged the glycemic responses of 800 people in response to nearly 50,000 meals; adding information about the subjects' microbiomes enabled more accurate glycemic response predictions. Since then, Golan said, additional trials have been conducted, most recently with the Mayo Clinic, to duplicate the results, and other studies are ongoing whose results have not yet been published.
He also pointed out that the microbiome was merely one component that goes into building a client's profile, in addition to medical records, including blood glucose levels. (I provided my HbA1c levels, a measure of average blood sugar over the previous several months.)
"We are not saying we want to improve your gut microbiome. We provide a dynamic tool to help guide what you should eat to control your blood sugar and think about combinations," he said. "If you eat one thing, or with another, it will affect you in a different way."
Viome acknowledged that the two companies are taking very different approaches.
"DayTwo is primarily focused on the glycemic response," Naveen Jain, the CEO, told me. "If you can only eat butter for rest of your life, you will have no glycemic response but will probably die of a heart attack." He laughed. "Whereas we came from very different angle – what is happening inside the gut at a microbial level? When you eat food like spinach, how will that be metabolized in the gut? Will it produce the nutrients you need or cause inflammation?"
He said his team studied 1000 people who were on continuous glucose monitoring and fed them 45,000 meals, then built a proprietary data prediction model, looking at which microbes existed and how they actively broke down the food.
Jain pointed out that DayTwo sequences the DNA of the microbes, while Viome sequences the RNA – the active expression of DNA. That difference, in his opinion, is key to making accurate predictions.
"DNA is extremely stable, so when you eat any food and measure the DNA [in a fecal sample], you get all these false positives--you get DNA from plant food and meat, and you have no idea if those organisms are dead and simply transient, or actually exist. With RNA, you see what is actually alive in the gut."
More contradictory food advice from Viome (left) and DayTwo.
Note that controversy exists over how it is possible with a fecal sample to effectively measure RNA, which degrades within minutes, though Jain said that his company has the technology to keep RNA stable for fourteen days.
Viome's approach, Jain maintains, is 90 percent accurate, based on as-yet unpublished data; a patent was filed just last week. DayTwo's approach is 66 percent accurate according to the latest published research.
Natasha Haskey, a registered dietician and doctoral student conducting research in the field of microbiome science and nutrition, is skeptical of both companies. "We can make broad statements, like eat more fruits and vegetables and fiber, but when it comes to specific foods, the science is just not there yet," she said. "I think there is a future, and we will be doing that someday, but not yet. Maybe we will be closer in ten years."
Professor Walter wholeheartedly agrees with Haskey, and suggested that if people want to eat a gut-healthy diet, they should focus on beneficial oils, fruits and vegetables, fish, a variety of whole grains, poultry and beans, and limit red meat and cheese, as well as avoid processed meats.
"These services are far over the tips of their science skis," Arthur Caplan, the founding head of New York University's Division of Medical Ethics, said in an email. "We simply don't know enough about the gut microbiome, its fluctuations and variability from person to person to support general [direct-to-consumer] testing. This is simply premature. We need standards for accuracy, specificity, and sensitivity, plus mandatory competent counseling for all such testing. They don't exist. Neither should DTC testing—yet."
Meanwhile, it's time for lunch. I close out my Viome and DayTwo apps and head to the kitchen to prepare a peanut butter sandwich. My gut tells me I'll be just fine.
AI was integral to creating Moderna's mRNA vaccine against COVID.
But the truth is that before ChatGPT nabbed the public’s attention, AI was already here, and it was doing more important things than writing essays for lazy college students. Case in point: It was key to saving the lives of tens of millions of people.
AI-designed mRNA vaccines
As Dave Johnson, chief data and AI officer at Moderna, told MIT Technology Review‘s In Machines We Trust podcast in 2022, AI was integral to creating the company’s highly effective mRNA vaccine against COVID. Moderna and Pfizer/BioNTech’s mRNA vaccines collectively saved between 15 and 20 million lives, according to one estimate from 2022.
Johnson described how AI was hard at work at Moderna, well before COVID arose to infect billions. The pharmaceutical company focuses on finding mRNA therapies to fight off infectious disease, treat cancer, or thwart genetic illness, among other medical applications. Messenger RNA molecules are essentially molecular instructions for cells that tell them how to create specific proteins, which do everything from fighting infection, to catalyzing reactions, to relaying cellular messages.
Johnson and his team put AI and automated robots to work making lots of different mRNAs for scientists to experiment with. Moderna quickly went from making about 30 per month to more than one thousand. They then created AI algorithms to optimize mRNA to maximize protein production in the body — more bang for the biological buck.
For Johnson and his team’s next trick, they used AI to automate science, itself. Once Moderna’s scientists have an mRNA to experiment with, they do pre-clinical tests in the lab. They then pore over reams of data to see which mRNAs could progress to the next stage: animal trials. This process is long, repetitive, and soul-sucking — ill-suited to a creative scientist but great for a mindless AI algorithm. With scientists’ input, models were made to automate this tedious process.
“We don’t think about AI in the context of replacing humans,” says Dave Johnson, chief data and AI officer at Moderna. “We always think about it in terms of this human-machine collaboration, because they’re good at different things. Humans are really good at creativity and flexibility and insight, whereas machines are really good at precision and giving the exact same result every single time and doing it at scale and speed.”
All these AI systems were in put in place over the past decade. Then COVID showed up. So when the genome sequence of the coronavirus was made public in January 2020, Moderna was off to the races pumping out and testing mRNAs that would tell cells how to manufacture the coronavirus’s spike protein so that the body’s immune system would recognize and destroy it. Within 42 days, the company had an mRNA vaccine ready to be tested in humans. It eventually went into hundreds of millions of arms.
Biotech harnesses the power of AI
Moderna is now turning its attention to other ailments that could be solved with mRNA, and the company is continuing to lean on AI. Scientists are still coming to Johnson with automation requests, which he happily obliges.
“We don’t think about AI in the context of replacing humans,” he told the Me, Myself, and AI podcast. “We always think about it in terms of this human-machine collaboration, because they’re good at different things. Humans are really good at creativity and flexibility and insight, whereas machines are really good at precision and giving the exact same result every single time and doing it at scale and speed.”
Moderna, which was founded as a “digital biotech,” is undoubtedly the poster child of AI use in mRNA vaccines. Moderna recently signed a deal with IBM to use the company’s quantum computers as well as its proprietary generative AI, MoLFormer.
Moderna’s success is encouraging other companies to follow its example. In January, BioNTech, which partnered with Pfizer to make the other highly effective mRNA vaccine against COVID, acquired the company InstaDeep for $440 million to implement its machine learning AI across its mRNA medicine platform. And in May, Chinese technology giant Baidu announced an AI tool that designs super-optimized mRNA sequences in minutes. A nearly countless number of mRNA molecules can code for the same protein, but some are more stable and result in the production of more proteins. Baidu’s AI, called “LinearDesign,” finds these mRNAs. The company licensed the tool to French pharmaceutical company Sanofi.
Writing in the journal Accounts of Chemical Research in late 2021, Sebastian M. Castillo-Hair and Georg Seelig, computer engineers who focus on synthetic biology at the University of Washington, forecast that AI machine learning models will further accelerate the biotechnology research process, putting mRNA medicine into overdrive to the benefit of all.
This article originally appeared on Big Think, home of the brightest minds and biggest ideas of all time.
