Sponsored By
alt text

Nutrition braces for impact from artificial intelligence and machine learnersNutrition braces for impact from artificial intelligence and machine learners

The Obama administration and other voices bringing genomics and microbiomics into louder discourse have primed the public for personalized medicine; now it’s time to deliver.

Marc Brush

December 3, 2016

10 Min Read
Nutrition braces for impact from artificial intelligence and machine learners

The ascent of Donald Trump casts a dark shadow on Obamacare and the lasting impact of any reforms the President brought to U.S. healthcare, but that shadow stops short in one important corner of medicine. Obama put personalization on the map with his 2015 State of the Union address, and pinned it in place with the White House’s subsequent Precision Medicine Initiative to fund research throughout the major science agencies in the US government. The ramifications of this are already being felt by progressive players in nutrition. “Personalization only got hot in the past year, after Obama made his announcement,” says Michael Nova, MD, Chief Innovation Officer at Pathway Genomics. “Now I’m giving presentations to insurance providers about artificial intelligence. That never happened before.”

But where are we with the research? Are supplements even part of the equation? It’s heady days for many business constituencies that touch personalized nutrition—from genetic testing labs like Pathway, to cutting-edge meal-kit providers like Habit, to high-end diagnostic platforms like Arivale—and heady days rarely last. Thanks to Obama’s vision for modernizing medicine, Vice President Joe Biden’s “cancer moonshot,” and countless voices bringing genomics and microbiomics into louder discourse, we’ve primed the public and set the groundwork, but now it’s time to deliver.

“I see huge promise in what big data can bring to scientific wellness and personalized nutrition, but we’re still in the early stages here,” says Nathan Price, Professor and Associate Director at the Institute for Systems Biology (ISB). (Price is also the co-founder of Arivale and a scientific advisor to Habit.) “There’s proof of concept in the marketplace now. We know that people will buy the devices, they’ll gather the data. As a field, we now need to jump to that next level and do something meaningful with it.”

How big is this data?

The pace of change in personalized medicine is quickening, and the landscape widening. The latest power player to plug in—machine learning and artificial intelligence—changes the conversation in radical ways, as big data begins to coalesce into big intelligence. “Big data is not the issue,” says Nova. “We’re trying to understand the data now. The issue is big understanding. Total healthcare data doubles every three years. That’s why machine learning is so important here.”

In 2014, IBM’s Watson group announced a partnership and equity investment in Pathway to bring its computational and deep machine learning to the forefront of wellness and quantified medicine. Out of this partnership, Pathway is set for a 2017 beta release of an app called OME that combines precision medicine and wellness with machine learning—a form of artificial intelligence that teaches computers to grow and learn without interventional programming by humans. This is important, because the big data of healthcare is immense, siloed, and more unruly by the day.

“By the year 2020, global healthcare data doubles every three days,” says Nova. “Each of us will generate two terabytes of healthcare data in our lifetimes, and that doesn’t even include DNA sequencing.” Given the current state of affairs, many of these data will come through unstructured, in no semblance of tabular form and largely unusable without the AI of machines like Watson. “The data is only as good as how it’s interpreted,” says Nova. “This is the perfect problem for machine learning and artificial intelligence. It’s intractable otherwise. You can’t overload AI with too much data.”

Price at ISB agrees. “We are gathering lots of information, but it’s not very well integrated. The interest now is in how much integrated information we can gather. We need apps to leverage that information and then reduce it down to a small number of action items that we can wrap our minds around and tie into behavioral coaching. Machine learning will be really valuable in taking advantage of this density of data, and I’m enthusiastic about it, but there’s still an important role for health coaches to enable those relationships and promote actionable behaviors.”

How smart is this intelligence?

So big data marches forward, and we begin to make sense of it in important ways, even without the therapeutic home runs many in nutrition see on the horizon. “More granular data and smarter machines are useful, but the really critical part is the intellectual input,” says John Mathers, Director of the Human Nutrition Research Centre at Newcastle University. “In other words, we need more work on the concept of personalization, on what factors influence how individuals respond to dietary advice, and on how to operationalize those concepts.”

NBJ asked a number of experts for examples of success in personalization, case studies wherein a genetic variant combines with blood markers to trigger lifestyle therapies that manage or prevent disease. The responses still lack the clarion call of victory, but success is closer.

Price speaks anecdotally of how the holistic pathways get built at ISB and how the dots begin to connect from big data to big intelligence. “Success can happen in a lot of places,” he says. Focusing on “scientific wellness,” he believes a lot of actionable possibilities already exist. As an example, he says, “an individual came through the ISB program with cartilage breakdown. He had arthritic symptoms in his knees, he didn’t know why, and the treatments weren’t helping. We did some advanced measurements, dove in on that, and what popped? His ferritin levels were super high, and his genome showed a variant associated with hemochromatosis. So he went back into the medical system for an easy treatment—bloodletting. Because he had the knowledge, a simple therapy removed an entire disease trajectory from his life.”

The anecdotes tend toward supplements as well. “Supplements are a mixed bag right now,” Price says, “but we believe that, when guided by measurements, targeted supplementation can be an important thing to do. A friend of mine some years ago wasn’t finding any zest in his work. He wasn’t excited about things, so he got a blood panel done, found severe iron deficiency, took some iron pills, and a few weeks later, he’s a whole new person. It’s a simple molecular issue, and it’s totally invisible.”

Food is an obvious focus as well. “Pre-diabetes is another example,” says Price. Doctors can monitor for insulin resistance and glucose responsiveness, and take actions to reverse the progression, but that’s not prevention. “We’d love to see what those pre-disease states look like, so we can take action before massive damage occurs.” Every diet is, in effect, personalized. A sedentary person can’t eat like an athlete. An endurance athlete doesn’t eat like a marathoner. “The question with food is how far does personalization go?,” Price says. “With measurements in hand of nutrient levels and toxins, the question becomes, can we tailor food so it’s more specific to the individual and more likely to produce health benefits downstream?” Price’s work with Habit, the meal-kit provider recently launched by Neil Grimmer, formerly of Plum Organics and Campbell, is designed to answer some of those questions.

State of the science

“A lot of the effort is still around gathering the data necessary to do the research,” says Price. “Studies diving in with lots of measurements are only now happening at any real scale.” Price is right. NBJ asked our sources for this article to point toward clinical trials with big data that reveal the way forward for personalized nutrition, and there’s just not a lot there yet. But there are signs of things to come.

Mathers at Newcastle University worked directly on the Proof of Principle Study at Food4Me, an EU-funded research effort into personalized nutrition. Mathers looked specifically at using the internet to deliver dietary advice to impact behavior in a large, complex study across seven European countries. “The Food4Me intervention study tested the hypothesis that a personalized nutrition approach would be more effective than a conventional ‘one size fits all’ approach,” says Mathers. He and his team tested three approaches to personalization: Level 1, based on analysis of diet alone; Level 2, based on diet plus phenotypic information, like anthropometry and blood measurements; and Level 3, based on diet plus phenotype plus genotype. “After six months, those people randomized to personalization had bigger improvements in eating patterns than those randomized to the control group with generic eating advice,” says Mathers. “Importantly, we also found that there was no additional benefit to including phenotype or genotypic information in the personalization approach.”

An oversimplified distillation would be that personalization works, but it’s too soon for the fancy stuff around genome and biomarkers. As Mathers puts it, “We’re still at an early stage in understanding the utility of genomic information in nutrition.”

Nova points to two studies of particular interest to the power of big data. Zeevi et al., appearing in the November 2015 issue of Cell, used a machine learner to predict the glycemic response of 800 people based on  data inputs from their microbiome and their genome, as well as food diaries, lifestyle questionnaires, and blood diagnostics. The machine learner created a “personalized nutrition predictor” that did, in fact, predict blood glucose responses and tailor diets that did lower post-meal blood glucose for better long-term metabolic health. Another distillation: AI wrangled a lot of the data into personalized diets that improved health. The machine could better predict who might flip into Type 2 diabetes over time and who might not.

Where it lands in nutrition

If the big data and smart machines are still crunching the algorithms to cure all known human disease, where might we start to see early successes in nutrition? Much of the interest at a consumer level revolves around diet and personalizing nutrition based on deeper blood diagnostics, while much of the interest at a patient level revolves around the prevention of food-related disease states like diabetes and obesity, not to mention some cancers and cardiovascular conditions. Somewhere in between lies a path to success for dietary supplements.

Food may continue to get the lion’s share of attention, given its impact on public health. “Poor nutrition and poor diet are the focus,” says Nova, “because they contribute to morbidity and mortality more than smoking and the next four risk factors combined.” Mathers sees the scale of the problem as well: “In my view, food preferences, psychological factors, and, indeed, fashions in food will influence how individuals choose to change their nutritional behaviors. From a health perspective, modifying intake of foods—as distinct from consuming supplements or functional ingredients—is likely to have a bigger and more pervasive impact on health and well-being.”

But there are certain areas gathering more attention than others among the science-minded players in nutrition. Cancer and precision medicine are now close bedfellows, and as we’ve seen above, diabetes research is benefiting from personalization tactics as well. The microbiome is a rich source of data tied inextricably to nutrition that’s getting more and more attention from researchers. To wit, Price focuses his research on all of the “omics,” but, when pressed, sees the most potential in the microbiome, the proteome (“you can address it by the genome”), and the metabolome (“a beautiful source of information that relates so much to function”).

No one can point to a specific condition that might benefit first and most as personalized medicine advances from data to intelligence, nor can they point to a specific micronutrient, but the machines are learning.

And they’re quick learners.

Subscribe and receive the latest updates on trends, data, events and more.
Join 57,000+ members of the natural products community.

You May Also Like