The couple struggled with stress and emotional pain. Frey was frustrated with how little was known about the causes of genetic diseases, and the lack of treatments.
So he decided to make a huge change in this life.
The painful episode was a trigger for a career metamorphosis that has brought Frey to a promising moment today. He’s launched a start-up, Deep Genomics, that thinks it can transform medicine by applying the hot field of deep learning to genomics.
Back in 2003 Frey had been researching computer vision at the University of Toronto, a hotbed of deep learning research. Deep learning is a type of artificial intelligence in which computers learn to identify and categorize patterns in huge data sets. For examples, a self-driving car could use deep learning to identify pedestrians. Or a photo app might automatically group all of your photos of your grandmother or beach vacation together.
With frustration fresh on his mind, Frey knew he wanted to work on genomics. Out of hardship, Frey sought a way to make a positive impact, and his deep learning background would be his secret weapon.
“How can I make a difference to the next couple that shows up for genetic counseling and needs to figure out what’s going on genetically,” Frey said. “A billion dollars had been spent on the genome, it was obviously very important, but really people could not make sense of the genome.”
Frey began to develop a rare skill-set — combining genomics and deep learning. He published papers in leading academic journals and gave talks at conferences and universities. But years went by and medicine wasn’t transforming. Other researchers and companies weren’t running with the work, which surprised Frey.
“It wasn’t enough to be publishing these research papers, we needed to get out there,” Frey said. “Not just publishing research papers that people are impressed with, but don’t know how to act on them, but actually getting inside of the value chain.”
His team would spin out of the University of Toronto and launch a start-up —Deep Genomics.
Frey had gradually realized how rare his expertise in both deep learning and genetics was. Deep learning is currently a hot field in technology, but it’s a small one.
Much of the talent has passed through the University of Toronto’s deep learning group. For example, Yann LeCun now leads Facebook’s artificial intelligence research lab. Geoffrey Hinton, who Frey had worked on machine vision with, now works at Google. Yoshua Bengio is collaborating with IBM’s Watson group.
Google and Facebook are paying engineers proficient in deep learning upward of $250,000 a year — and the most experienced can earn in the seven figures, according to a Re/code report.
While most deep learning work tries to bring human capabilities to machines, Frey wants the machines to be able to do something humans can’t do.
“Humans are good at text analysis, humans are good at speech recognition, but humans do not understand the text of the genome,” Frey said. “So it’s a much more exciting problem to me, it’s like we’re completely in the dark.”
A deep learning system that’s able to digest a massive amount of genetic data has the potential to understand the impact of genetic mutations better than humans ever have.
The impact of mutations depends on their context. As Frey gets access to more data sets, of say individuals with autism, the deep learning system can better draw conclusions about how genetics is driving real world outcomes.
For now, much work remains. LeCun described Frey’s progress so far as a baby step. But the huge potential is there.
LeCun recalled to me watching Frey give a talk about his work in December to a group of deep learning researchers.
“We all thought it was amazing, and a pretty great tool once the dust settled,” said LeCun, who is now on Deep Genomics board of directors. He also cautioned about legal and commercial obstacles.
The long-term goal for Deep Genomics is to transform medicine through pharmaceuticals and personalized medicine. But Frey, aware of the regulatory and clinical trials that make pharmaceutical work a slow process, is targeting genetic testing in the short term.