Genetics can predict height, cancer risks, neuropsychiatric diseases and more — but not IQ?
Is a genetic analysis for IQ snake oil? Complex, yes. Unscientific? No.
In the East Room of the White House, Francis Collins, one of the pioneers of the Human Genome Project, unveiled the first draft of a complete sequence of human DNA. The year was 2000.
“We have caught the first glimpse of our own instruction book, previously known only to God,” he uttered.
One of the greatest technological achievements ever, the Human Genome Project is the foundation of modern genomics, heralding the promise of a future where we understand the human blueprint.
The message was one of optimism, an eagerness for an era of application, where this new draft of the human blueprint could transform our lives.
Unfortunately, in the last 20 years, that’s changed. When genetics enters the public view today — most often by way of products like Nucleus, which I founded in 2021 — researchers and officials take a different, precautionary tone.
In October, Nucleus Chief Scientific Officer Lasse Folkersen and I provided early interviews to Undark, a science magazine published with funding from MIT. The article leveled sweeping, misleading criticisms of new DNA analyses and falsely implied that Nucleus hopes to recklessly evade regulatory scrutiny. This month, when Nucleus launched a waitlist for Nucleus IQ, our genetic analysis for intelligence, we were accused of selling entertainment as science.
Though academia and industry don’t always see eye to eye, I appreciate that both want to bring critically important conversations at the forefront of genomics to the forefront of society.
But I want to draw attention to the fundamental question at stake: Who decides what you get to learn from your own DNA? I believe these insights — with clear limitations and context — belong in the hands of the public.
As such, Nucleus isn’t the new intelligence score company, nor is it another shady player in what Undark described as the “murky market” of genetic prediction.
We are the most affordable and accessible clinical-grade genetic test in the U.S. that reads and analyzes all your DNA. It’s a platform where anyone can access state-of-the-art genomic insights to learn all that there is to currently know about how their DNA shapes them.
Recently, I’ve gotten a lot of questions about Nucleus IQ. So let’s use our intelligence analysis as a lens into our approach.
Is IQ genetic?
Sasha Gusev, an associate professor of medicine at Harvard Medical School, wrote about this last weekend, underscoring a growing debate among scientists: What can a genetic model for IQ predict today — and what can it predict in the future?
From Gusev’s Genomic prediction of IQ is modern snake oil:
“With the magic of statistics, we can already put an upper bound on what the best prediction accuracy can be. This is the estimated “SNP-heritability”...In Tanigawa et al. 2022, the average SNP heritability was 17%...indicating that even with all of the training data in the world, a predictor built from common variant GWAS will end up in the modest but potentially useful range for the typical trait.”
But is SNP heritability really the upper bound of genomic prediction? No, it isn’t.
SNPs, or single nucleotide polymorphisms, are common genetic differences (or “variants”) that each make small contributions to a biological outcome. As a result, SNP heritability measures the collective role these kinds of variants play in something we can observe — in this case, IQ.
There’s also another measure of heritability called broad-sense heritability, which isolates genetic influence from the influence of one’s environment. Scientists compare identical twins, who are genetically identical, to fraternal twins, who are roughly as similar genetically as typical siblings. The idea is that stronger similarities between identical twins can be attributed to genetics, showing DNA plays a larger role in an outcome, while differences signal other factors at play.
Unlike studies of SNP heritability, twin studies have found surprisingly similar measures of IQ among identical twins, suggesting IQ is predominantly genetic with a broad-sense heritability of up to 80%. Even studies of identical twins raised completely apart found 70% of IQ was genetic.
Twin studies may overshoot the influence of genetics on IQ, but there are strong reasons to believe it’s substantially higher than the 19% suggested by SNP heritability — especially because SNP heritability makes overly simplistic assumptions about genetic influences.
In fact, scientists have suspected for almost 20 years that we’ve systematically ignored the effects of complex genetic interactions by making two crucial assumptions about how DNA shapes us. The first is that common SNPs are the only kind of genetic variants that contribute to biological outcomes, and the second is that these variants have independent, linear, and purely additive effects.
More precise genetic prediction will require models that can capture the complex interactions between common SNPs and rare genetic variants, which are DNA differences that can strongly shape individual biological outcomes.
Increasingly sophisticated genetic models are already capturing presumably more complex genetic influences using machine learning — even resulting in models that outperform traditional ones by up to 100%. Already, researchers are employing such models for common diseases like Alzheimer’s disease and breast cancer with strong early results.
Now, what about rare variants? Uncovering those will take more data spanning the entire genome, not just the fraction that old technology could capture. Sequencing that cost millions of dollars for decades — fundamentally limiting our understanding of genetic variation — is now a few hundred dollars.
On a population level, rare variants — the kind you can garner from sequencing DNA — might not look like much. In some cases they may even seem to amount to barely a rounding error when it comes to predictive power. But on an individual level, these rare variants can be excellent predictors. For example, it was a rare variant in the BRCA1 gene that helped increase Angelina Jolie’s risk for breast cancer from 13% to almost 90%.
What would a variant like that mean for IQ? Well, it would be a profound discovery that would fundamentally shift our understanding of the genetic basis of intelligence. Not only would it help establish the molecular basis of intelligence, but also serve as an individual predictor that has much higher certainty.
So how accurate are today’s genetic IQ predictions? They’re not snake oil — they’re a starting point.
Today’s genetic models of IQ can’t tell you your IQ. No genetic model will ever be able to do that. Scientists will also never identify a single gene for intelligence. Instead, these analyses begin to tell you something much more fluid (and interesting, in my opinion): how your genes (nature) stack up against life (nurture).
Pitting nature against nurture is a false dichotomy. The answer always lies somewhere in between. In the case of intelligence, the contribution of genetics can be nearly eliminated by lack of educational resources and poverty. It is also true that IQ is predominantly genetic without these social constraints.
In other words, DNA is never destiny — even once we achieve a perfectly predictive model.
In scientific terms, today’s models — including Nucleus IQ, which predicts your ability to reason or solve problems across a few IQ points — are modest predictors of intelligence.
Some scientists, including Gusev, have suggested selling genetic IQ prediction as entertainment. But does that really make sense?
Genetic models inevitably improve over time. Today’s models for height can predict your height within a few centimeters. Similarly, models for IQ will continue to improve over time. Models for height didn’t start at today’s accuracy. Science doesn’t begin at the finish line.
Moreover, genetic analyses for IQ also have the same underpinning as more widely-accepted clinical use cases. Genetic models today can effectively identify your risk for some of the most common diseases in the U.S., like heart disease, type 2 diabetes, and breast cancer. Combining these predictions with lifestyle factors, like someone’s age, sex, height, and weight, can make them even stronger.
Certainly, some genetic models have low predictive power. Models for gastric cancer, depression, and skin cancer in fact, have a lower predictive power than the best analyses measuring intelligence. But we wouldn’t consider these entertainment. We’d consider them the evolving frontier of our understanding.
The future of genomics
Why was Shawn Bradley almost 3 feet taller than the average man? How was it possible for Richard Feynman to master differential and integral calculus by age 15 — and was it coincidence that his sister, Joan, was the astrophysicist who discovered the origins of auroras? Why did chronic illness — ranging from stroke to Addison’s disease — plague the Kennedy family?
Ultimately, genomics helps us answer questions about human variation. As more precise models of heritability combine with machine learning, scientists and practitioners will be able to predict genetic influences with greater accuracy.
The advent of widespread whole-genome sequencing will benefit each of us, opening access to the complete range of scientific and medical knowledge into how genetics shapes who we are. That’s why Nucleus exists.
 | A guest post by
|