Silicon Valley thrives on acceleration. PG said it best back in 2012: Startup = growth. The ambition of every generational company is to, as quickly as possible, productize the frontier of science and technology and make that innovation available to as many people as possible.

For genomics, this frontier is the ability to read and interpret DNA. In the last two decades alone, the cost of measuring nearly every genetic marker to capture their impact on our lifelong predispositions to every possible disease has dropped from $20 million to $200.

Our ability to read and understand the genome is inextricably linked to our ability to engineer it. My belief in this dates back to high school, where my interests didn’t yet lie in computational genomics, but in genetic engineering.

I’ve been thinking about how the “genomic stack” — or the way humanity will neatly integrate and productize all the pieces of genomic technology — ever since then.

Genetic engineering will undoubtedly be one of the core applications. But the way tech talks about engineering humanity today is a bit like if Kennedy shot for Mars before trying to land on the moon. We don’t actually understand the genetics of human intelligence well enough to engineer it in babies. 

Take Nucleus IQ, for example. We analyze over 500,000 markers that are known to be associated with intelligence to predict whether someone is predisposed to having, on average, a slightly higher or lower IQ. 

So, you might assume that the best way to engineer intelligence is to edit each of these 500,000 markers that we know correlate with intelligence. 

Not so fast.

Each of these 500,000 associations denote regions of DNA that are associated with intelligence. But we don’t know the specific genetic markers that actually drive intelligence.

It’s like correctly predicting the NBA championship winner based on the team’s regular season record — without knowing which players actually drove the team’s success and why. Is that cool? Yeah. But could we coach the team to a win? We couldn’t.

You can see the pattern, but you don’t understand the cause. 

In the same way, we can predict human intelligence (albeit with wide error margins) — but we can’t engineer it. We might eventually uncover the 500,000 markers that contribute to IQ to engineer intelligence, but that’s not where we are today. Even then, actually editing 500,000 genetic markers is still a tall task.

So where does that leave us?

We shouldn’t throw cold water on the idea of gene editing. Instead, we should come up with a tactical plan to actually do so.

An essential law of genetics is that the impact of a genetic marker is inversely correlated with its frequency. In other words, even if we could engineer one of the 500,000 causal markers linked to IQ today, it would have a negligible effect on someone’s IQ. This is because these are all common markers found in the population and common markers tend to have very small effects on any biological outcome.

That also means that rare genetic variants have an outsized impact on genetic outcomes, making them excellent individual predictors — and excellent candidates for genetic engineering. Rather than coming down to the subtle impacts of possibly millions of genetic markers, it’s much more likely that intelligence — like the most well-studied chronic diseases today — can be strongly stratified by rare genetic variants that we historically hadn’t had the technology to uncover.

Think of breast cancer, one of the most well-studied chronic diseases today. 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 on what the IQ of the baby would be. 

Overindexing on common genetic markers in the context of genetic engineering ignores the fact that there is likely a single high-effect genetic marker that dramatically increases someone’s genetic predisposition for an outcome. Discovering those genetic markers has just become possible in the last decade with the proliferation of cheap whole-genome sequencing.

The best genetic engineering startups know this. Why? Simple — single corrections in DNA are far easier to engineer than hundreds of thousands.

Whole-genome sequencing helps us understand the source of genetic variation in human beings. This roadmap — identifying this single genetic marker and then engineering it — is a great first step for illustrating the power of genomic engineering technology while minimizing tradeoffs.

So next time you read an article about engineering 500,000 markers in DNA — know there’s an easier way.

Fun article theorizing about hereditary polygenic editing. I agree very much with the ethos of acceleration, though some of the points on polygenic editing today are not as sound. Inspired me to write this newsletter!

Last week, NVIDIA and research nonprofit Arc Institute released a deeply trained AI model that can predict what yet unknown genetic markers do — and design new DNA sequences. Biology is accelerating.

Masterful reporting in Core Memory on The Los Angeles Project by Josie Zayner and Cathy Tie. Decades of genetic research is going unused by academics. Tech is seizing the opportunity.

On February 18, Humane, once a hyped AI hardware startup in Silicon Valley, was partially acquired by HP for $116 million — less than half its $240 million venture capital haul, and after five years of secret development. The lesson? Startups and their products gotta touch grass. Any startup that isn’t launched should be written off as dead.

A section dedicated to the before-and-after of everyday creation.

Thank you for reading. If you enjoyed, forward to a friend so they can sign up here.