Silicon Valley’s murky push to predict your health with your DNA

To hear Kian Sadeghi tell it, his genetic testing company just might upend healthcare. The 23-year-old college dropout began working on Nucleus in 2020, from his bedroom in his parents’ Brooklyn home. From a swab of genetic material, he says, his company can glean insights into which drugs a person should be prescribed, which diet will best help them lose weight, how concerned they should be about getting Alzheimer’s, and many other health considerations.

Sadeghi has received millions of dollars in financial backing from some of the most influential figures in Silicon Valley, including venture capital firms led by Reddit cofounder Alexis Ohanian and billionaire Peter Thiel. “It’s an inevitability that every single person has their entire human genome on their iPhone,” Sadeghi told Ohanian in a promotional video released on X (formerly Twitter) in July 2022. “Who’s going to analyze it? Who’s going to store it? Who’s going to maintain it?” he said. “It’s going to be Nucleus.”

Sadeghi, who studied computational biology during his year-and-a-half as an undergraduate, asserted that he can provide more accurate, more extensive genetic analysis than anyone else in the industry. And as he described it, he is under no obligation to the U.S. Food and Drug Administration (FDA), the agency charged with ensuring the safety and efficacy of commercial genetic tests, to demonstrate that any of it works as well as he’s claimed.

In that regard, he is not alone. Nucleus is part of a brewing industry that seeks to broaden the scope of consumer genetic testing—to move beyond the mere detection of rare genetic variants known to cause serious disorders and into a more slippery realm of probabilities. They capitalize on tools that mine huge DNA databases in search of genetic signatures for a virtually endless range of human characteristics: the odds that a person will develop a certain cancer, for example, or the likelihood that they’ll go far in school or obtain a high-paying job.

The companies commercializing these techniques have largely escaped regulatory scrutiny, despite doubts about the health benefits their analyses can provide and worries about the risks they could pose to customers, healthcare systems, and society at large. The commercial push has created tension between academic researchers who toil to uncover new genetic associations, and the entrepreneurs who repurpose the results for profit.

“It’s an inevitability that every single person has their entire human genome on their iPhone.”

Some ethicists also express a darker anxiety: a worry that companies will be tempted to test the blurry line that separates genetic prediction of disease risk from attempts to divine more complex traits laden with social stigmas. They portray it as a slippery slope that could lead to new modes of discrimination and bias.

Whether Sadeghi himself will test that line remains unclear, but the 23-year-old CEO has hired as his chief scientific officer Lasse Folkersen, who once ran a site offering genetic scores for intelligence and other scores seen as controversial in genetics research, taking extraordinary steps to shield some of the offerings from public view.

A sobering question emerges: Is the world ready to be awash in genetic scores? “No,” said Lucas Matthews, an assistant professor of psychiatry at Columbia University who has studied some of the technology’s social ramifications. “But I don’t know if it ever can be.”

“It’s already out there,” Matthews added. “It’s happening, whether we’re ready or not.”

In a March 2022 video interview with the 20 Something Podcast—a show exploring resonant topics for people in their twenties—Sadeghi gave listeners a preview of how Nucleus will work: “You just basically go on our website [and] if you haven’t done genetic testing before, you can just order a kit off of it.” Once a customer has collected a DNA sample and sent it back to Nucleus, he explained, the company then processes and analyzes it, unlocking a hidden world of personal genetic insights. “We have this beautiful kind of disease and trait panel,” he said. “You can click on any disease.”

Dressed plainly in a black T-shirt with his bedroom as “backdrop,” Sadeghi then reeled off a half-dozen or so examples of the many diseases and traits Nucleus will analyze: Alzheimer’s, cancer, depression, extroversion, and longevity among them. More analysis on more diseases and more traits, he promised.

The idea that someone could divine, from a DNA sample, a person’s odds of developing Alzheimer’s might seem like a fantasy. After all, there isn’t an “Alzheimer’s gene”—no single snippet of genetic code that determines if a person will develop the disease.

But in recent decades, scientists have come to appreciate that there are, in fact, many genes that each seem to have some bearing on a person’s Alzheimer’s risk. To identify them, they can perform what are known as genome-wide association studies—data-intensive efforts that trawl vast human genetic databases to tease out correlations between genetic variants and human characteristics. One recent study, which examined the genetic information of nearly 800,000 people, found 75 points on the genome that have genetic variants associated with changes in Alzheimer’s risk.

Genome-wide association studies, or GWAS, are not limited to Alzheimer’s. For seemingly any human characteristic that can be measured, there are GWAS that have identified potentially relevant genes. A database maintained by the National Human Genome Research Institute and the European Bioinformatics Institute catalogs more than 88,000 such studies that have probed more than 9,000 traits. Studies have found genetic variants correlated with a tendency to develop diseases, like coronary artery disease and diabetes. Others have found correlations between genes and complex outcomes like the age at which a person loses their virginity, or their propensity for taking risks.

The techniques have become so routinized that even an automated bot can perform them: For the past five years, GWASBot, maintained by researchers in the U.S. and Finland, has conducted roughly one new GWAS every day, publishing its results on X. On different days in August, it identified genetic variants linked to people’s ability to solve an arithmetic problem, their tendency to feel fed up, and their penchant for taking naps during the day.

The biological mechanisms underlying the genetic associations are, almost always, opaque. Often, it is impossible to know whether a variant identified in a GWAS causes—or is merely correlated with—a change in the observed characteristic. Sometimes, the associations may not have much biological meaning at all: A specific point on the genome can be correlated with a trait for all sorts of reasons that have little to do with the gene itself.

A sobering question emerges: Is the world ready to be awash in genetic scores?

Despite that uncertainty, the GWAS boom has bred a prediction industry, of sorts. Anyone with enough technological savvy can take a list of the genetic variants identified in a GWAS—and their effective sizes—and cross reference it with a person’s genetic information to calculate a number, a so-called polygenic score, that reflects that person’s predisposition for a disease or trait. The score may be presented as a percentile ranking that describes the person’s standing compared to a broader population and—in some cases—as an absolute risk that estimates a person’s overall odds of developing the disease or trait.

The estimates are imperfect, in part because genetics aren’t the only factors that influence a person’s outcomes; environmental influences also matter. A 2022 study of Swedish twins found that polygenic scores for Alzheimer’s could explain about 11% of the variation in risk that researchers observed—even after specially accounting for a gene that encodes for the protein apolipoprotein E, thought to have an outsized influence on the disease. Much of the remaining variation was ascribed to lifestyle and other nongenetic factors. Scores for other traits paint a similar picture: A 2023 study found that polygenic scores for coronary artery disease explained about 7.4% of the observed risk variation in a population of European ancestry; in a 2018 study of breast cancer, the fraction was just 2.7%.”

Taken alone, the scores typically reveal very little about a person’s likelihood for developing a trait. In a 2019 study that followed people aged 60 and older who initially showed no signs of cognitive impairment, just over 1% of people whose polygenic scores put them at relatively high Alzheimer’s risk went on to develop dementia during the study’s roughly 8-year observation period. Just 0.63% of people who scored at the lowest level of risk developed dementia in that same timeframe. The vast majority of people, whether their genetic risk was deemed high or low, didn’t develop the condition.

A specific point on the genome can be correlated with a trait for all sorts of reasons that have little to do with the gene itself.

Although polygenic scores alone may not be powerful predictors of disease, many researchers are optimistic that they can be combined with other, more conventional methods of risk estimation to improve screening for common conditions like cancer, cardiovascular diseases, and diabetes.

In the United Kingdom, the healthcare company Genomics partnered with the National Health Service to integrate the scores into traditional risk calculators for cardiovascular disease based on factors like cholesterol levels and blood pressure. A study funded by the company found that estimates that combined polygenic scores with those traditional risk factors were about 6% more accurate than those produced with the traditional risk calculator alone. They predicted that around 2,400 deaths could be prevented annually in the U.S. if the modified risk tool were implemented, and if every person identified as high risk adopted and adhered to a regimen of drugs that reduce cholesterol, called statins.

Other studies, which were not funded by industry, have found more modest gains. But collectively, they have fueled optimism that clinicians may one day use the scores to detect some ailments earlier and more accurately, empowering preventative measures that could help patients ward off disease and serious illness.

At the same time, however, a chorus of experts say that society should temper any expectations that polygenic scores will revolutionize healthcare. Among them is Amit Sud, who previously served as an academic clinical lecturer at the Institute of Cancer Research. In recent papers, Sud and coauthors have pointed out that the accuracy gains from using polygenic scores are modest and that, as with any screening tool, the possibility that a polygenic score will correctly flag a candidate for preventative treatment must be balanced against its potential for false positives—cases where someone is flagged as high risk for a disease that they will never develop.

Those false positives, Sud and his colleagues have argued, could lead many people to undertake costly and unnecessary screening and treatment. In an analysis published this year in The BMJ, they estimated that a proposed U.K. policy to offer annual screening to women with moderate or high-polygenic risk scores for breast cancer would detect 1,700 more cancers but produce more than 5,700 false positives, and they pointed to a separate study estimating that a proposal to incorporate polygenic scores in coronary artery disease prevention would cost around $140,000 in additional screening and treatment for every year of quality life that it saved.

To the extent that clinical benefits of polygenic scores do materialize, Sud and his coauthors argued, they are unlikely to be reaped equally. That’s because, for technical reasons, the scores tend to work best for people who are genetically similar to the populations studied in the genome wide association studies the scores were based on. And, to date, the vast majority of GWAS have used populations of European descent.

“When you hear people speak about polygenic scores,” he added, “it sounds very optimistic. It sounds like it’s going to solve a lot of problems. But from first principles, I don’t see how it can.”

If the benefits of introducing polygenic scores in controlled clinical contexts remain uncertain, some say there’s even less evidence to suggest that individual consumers will be able to reliably parlay scores into improved health outcomes on their own. Yet improved health outcomes are exactly what Nucleus and many other consumer-driven genetic companies are promising.

“I think it ultimately comes down to precision personal insight,” Sadeghi said when he was asked in the March 2022 podcast interview what actionable steps customers would be able to take based on the information that Nucleus provides. “What drugs should be used specifically for you, in the health care context? What diet should you have based off your DNA?”

“When you hear people speak about polygenic scores, it sounds very optimistic. It sounds like it’s going to solve a lot of problems. But from first principles, I don’t see how it can.”

Other companies have implied that they can deliver similar benefits. Nebula Genomics, a genetic testing and analysis company cofounded by Harvard University professor George Church, advertises that its analyses can help customers determine the best diet and supplements, find the right exercise plan to lose weight, and extend their lives. The company charges $249—plus a $275 lifetime membership fee—for its whole genome sequencing package, which includes reports on ancestry and rare genetic mutations, as well as access to nearly 290 polygenic scores on different genetic predispositions.

However, Cathryn Lewis, a professor of genetic epidemiology and statistics at King’s College London, said she’s seen scant evidence that people react to their polygenic scores in ways that yield positive impacts on their health. She said the closest study she knows of is a 2022 randomized controlled trial that tracked patients with high-polygenic risk for coronary artery disease.

The study found that patients who were informed and counseled about their polygenic scores were more likely to begin a regimen of statins, compared with patients who weren’t apprised of their scores. But after one year, the two groups showed no appreciable difference in their risk for cardiovascular disease, based on factors like blood pressure, cholesterol levels, and blood glucose levels. Learning about polygenic scores changed people’s behavior, Lewis said, but it didn’t have the intended effect on people’s health.

Absent clearer evidence of polygenic scores’ utility, some experts are leery about efforts to position them as consumer health products. In an email to Undark, Muin Khoury, the chief of the public health genomics branch at the Centers for Disease Control and Prevention (CDC), was blunt: There are no polygenic score tests, he wrote, “currently available for health, social, or behavioral traits with sufficient scientific evidence to support their routine use in health practice.” Absent guidance from a healthcare professional, he added, “a consumer is on their own to determine whether a genetic test is valid and useful for examining health and other traits.”

Lewis told Undark that she worries the analyses will send customers rushing to their primary care providers to request unnecessary screening and tests. And because people with the means to purchase the analyses are likely to be more affluent and middle class, she added, the products could exacerbate inequalities in access to care. “It rolls into all sorts of downstream medical care,” Lewis said, “which is not appropriate for the individual. It’s not appropriate for society.”

Church acknowledged that many polygenic scores have not demonstrated an ability to deliver clinically significant health outcomes, but he indicated that Nebula Genomics’ reports are not intended for medical or pharmaceutical use, and that “the point of the exercise is to get people used to thinking probabilistically.” He said he hopes they will urge consumers to talk with their doctors and seek more diagnostics, and he suggested that fears that those patients will strain healthcare systems are overblown. “I think we complain too much about medical costs,” he said. “I think we’re underdiagnosed as a species, given the amazing technology.”

Even among some researchers who are optimistic about using polygenic scores to screen for physical health conditions, there is one emerging application of polygenic scores that makes them uneasy: the prediction of risks for depression and other psychiatric conditions.

The genetic testing giant 23andMe has helped lead the way into this ethically fraught terrain. In recent years, the company has released more than three dozen health-related polygenic scores, including scores for depression, ADHD, and panic attacks, according to company white papers. Meanwhile, Nebula Genomics’ offerings include scores for depression, PTSD, bipolar disorder, and other mental health conditions. And although Nucleus has yet to disclose its full genetic testing lineup, Sadeghi has already indicated that depression will be on the menu of options.

For people like Lewis, who herself has helped conduct several genome-wide association studies on depression, part of the concern is that geneticists’ understanding of psychiatric traits is still immature. In a widely cited 2018 study, she and her coauthors found that polygenic scores could account at best for just about 2% of the variation in depression risk observed in general populations—though she says that figure has been as high as 6% in unpublished data. Still, “giving you information on that very small piece of information isn’t very useful when the rest of it is unknown,” said Lewis.

“I think we’re underdiagnosed as a species, given the amazing technology.”

Experts also worry about the psychological harms that might come from receiving a high-risk determination for a disease like depression, with no clear course of action to take in response. “If so