Brendan Frey thinks his company could solve what many consider the biggest problem in genetics, the thing that keeps us from from using our genetic blueprint in order to completely transform the way we understand our health.
The problem is that despite our rapidly improving ability to quickly read and sequence a human genome and despite our even more astounding ability to accurately and cheaply edit the genetic code, we don't know what most of the genome actually says.
We don't yet know how a complex trait like intelligence is explained by our genes. Diseases like Alzheimer's are at least partially explained by genetics, but we're still trying to figure out which genes are involved. The same is true for many other illnesses.
The effects of a particular random genetic mutation, which happen all the time but may have a great effect or may have no effect, are even more unknown.
"After 110 years of genetics, and 15 years after the $3.8 billion Human Genome Project promised fast cures, after more billions spent and endless hype about results just around the corner, we have few cures," David Dobbs wrote in May, in a much-discussed story for Buzzfeed. "And we basically know diddly-squat."
We can see and change billions of letters, but we simply don't know what many of them mean.
Frey's startup, Deep Genomics, is leveraging artificial intelligence to help decode the meaning of the genome, something he started investigating in his lab at the University of Toronto. Specifically, the company is using deep learning: the process by which a computer takes in data and then, based on its extensive knowledge drawn from analyzing other data, interprets that information.
Their learning software is developing the ability to try and predict the effects of a particular mutation based on its analyses of hundreds of thousands of examples of other mutations — even if there's not already a record of what those mutations do.
They're trying to build not just a Rosetta Stone that explains a language, but a way to predict how a tiny change in the letters will create something new.
'The ultimate goal'
For Frey, the inspiration behind the company was personal.
Back in 2002, Frey's wife was pregnant. Tests revealed that there was something wrong with the baby, caused by some sort of genetic problem.
But doctors and genetic counselors had no idea what exactly the problem was.
"Living with that uncertainty was exhausting — diagnosticians couldn't make sense of it," Frey tells Tech Insider. "That was really frustrating."
It was hard to hear that doctors knew the problem was somehow explained by genetics but couldn't read that code well enough to pinpoint what was actually happening. Further, he says, he realized that most researchers weren't even trying to understand the full text of a human genome.
"Everyone agreed that being able to understand how DNA generates life is the ultimate goal," Frey says, "but they were pretty pessimistic [about being able to do it], so they didn't try." Jumping from being able to sequence those billions of letters to actually interpreting them like a language was too intimidating.
Using genetics in medicine at that time usually involved seeing if doctors could match an existing problem with a known mutation, Frey explains. But that didn't help answer questions about unknown problems.
Frey specialized in machine learning at that time — not genetics. His work was building computers that could learn.
Based on his experience with artificial intelligence and deep learning, he thought he could perhaps (with the help of people who knew more about genome biology) design a way to help a machine learn to interpret the genetic code.
What they still need to figure out
So far, Deep Genomics has used their computational system to develop a database that provides predictions for how more than 300 million genetic variations could affect a genetic code. They've published results from their work in the journal Science, explaining how this has led to new insights into the genetic connections of autism, cancer, and spinal muscular atrophy. That predictive system has revealed things about the genome that weren't known before.
But as Frey explains, it's still not perfect and far from complete. Right now, he says, "there could be a mutation that our system predicts is going to be problematic," but while that mutation might have an effect — which is part of what the system is predicting, since most mutations don't do anything — "it's just not a disease." That would lead to many results that are something like a false positive.
The system needs to learn more and get better at interpreting genomic data. Right now, it's still building the tools to find meaning in all that mysterious code.
As a company, Deep Genomics is still getting started. Frey explained that they hope to grow from five full-time employees to 12 soon, so they can start working on a bigger scale.
There are tons of companies out there doing fascinating work with genetics and gene sequencing especially, Frey says. But he argues that they're doing something no one else is.
"If you look at what's out there in terms of this field, there's quite a few companies that have emerged," he says, especially companies working in gene sequencing and others that are working on processing that data. But he says "you won't find any that say we're going to connect the genome to what you find in your cells." Having a system that analyzes mutations and predicts what effect they will have is still unique, Frey says.
If that system can predict the effect of mutations, that's something that will help doctors figure out diseases in ways that have never been done before. Combined with new gene editing technology, it could be the thing that helps transform how we use genetics in medicine.