Canada Tries to Turn Its A.I. Ideas Into Dollars

Featured in The New York Times, April 9, 2017

Deep Genomics, founded in 2014, has stayed in Canada, and its American-based venture backers encouraged it to remain in Toronto.

Brendan Frey, the chief executive, studied under Mr. Hinton at the University of Toronto, and he has spent years on research that combines deep-learning A.I. and cell biology. When he hires software engineers, he asks them to make multiyear commitments.

“There are a lot of distractions in the Bay Area,” said Mr. Frey, who is also a professor at the University of Toronto and a co-founder of the new Vector institute. “The hype is a little too hot down there. Besides, we have some of the best talent in the world here.”

How artificial intelligence could transform the medical world

Featured in The Star, May 9, 2016

Artificial intelligence is already powering your Google searches, your Netflix recommendations, and your smartphone’s virtual assistant. It is playing humans at complex, intuitive games like Go, and it is beating them.

Now, researchers say, they want AI to power your doctor’s diagnoses, your drug prescriptions, and your smartphone’s virtual psychologist. They want AI to perform tasks that radiologists do, and at least match them.

Machine learning has made tremendous strides in the last decade, becoming one of the fastest-growing, most-hyped areas of computer science. For the researchers who work at the intersection of health care and machine learning, the road ahead is steeper.

“I can say for sure that winning a game of Go is actually quite easy compared to understanding human health, and extending life spans, and saving lives,” says Brendan Frey, the co-founder and CEO of Deep Genomics and a professor of engineering and medicine at the University of Toronto.

Deep learning meets genome biology

Featured in O'Reilly, April 27, 2016

Genome biology, as a field, is generating torrents of data. You will soon be able to sequence your genome using a cell-phone size device for less than a trip to the corner store. And yet, the genome is only part of the story: there exists huge amounts of data that describe cells and tissues. We, as humans, can’t quite grasp all this data: we don’t yet know enough biology. Machine learning can help solve the problem.

At the same time, others in the machine learning community recognize this need. At last year’s premier conference on machine learning, four panelists—Yann LeCun, director of AI at Facebook; Demis Hassabis, co-founder of DeepMind; Neil Lawrence, professor at the University of Sheffield; and Kevin Murphy from Google—identified medicine as the next frontier for deep learning.

To succeed, we need to bridge the “genotype-phenotype divide.” Genomic and phenotype data abound. Unfortunately, the state-of-the-art in meaningfully connecting these data results in a slow, expensive, and inaccurate process of literature searches and detailed wetlab experiments. To close the loop, we need systems that can determine intermediate phenotypes called “molecular phenotypes,” which function as stepping stones from genotype to disease phenotype. For this, machine learning is indispensable.

Using deep learning to analyze genetic mutations

Featured in News Medical, Sept 21, 2015

We’re excited by the possibilities that DeepBind can open up for clinical and R&D work around the world. We and others have used DeepBind to identify mutations that disrupt protein binding within the context of disease.

Deep Genomics has an aggressive science and technology roadmap for building a computational system that links together many components that account for different cellular processes.

Think about the Google search engine, but for human mutations. Our first product, SPIDEX, is just one part of that system. In fact, SPIDEX already relies on a simple system that predicts where proteins will bind. We will replace that system with DeepBind and produce SPIDEX data that is much more accurate. All of these components interact and improving one of them boosts the performance of the others. That’s also how deep learning works in general.

We're finally cracking the secrets of what makes us sick

Featured in Tech Insider, Oct 26, 2015

Scientists all over the country are pushing for new ways to understand genomic data.

At 23andMe, researchers are collecting data from the more than one million people who have spit into a tube, sent their genetic material to the company to learn more about themselves, and consented to have their genetic information used for research.

Making sense of the links within such vast stores of data will require technologies that are only now becoming powerful enough to help.

Deep Genomics, a startup run by Brendan Frey, is leveraging artificial intelligence to help decode the meaning of the genome.

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.

Deep Genomics' learning software is developing the ability to 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 an as yet largely inscrutable body of text, but a way to predict how a tiny change in the letters will create something new.

Meet Deep Genomics, a start-up bringing the power of deep learning to genomics

Featured in the Washington Post, July 22, 2015

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.

Using deep learning and artificial intelligence to map the genome and predict disease

Featured in MedCity News, July 22, 2015.

Deep Genomics, a fresh new University of Toronto spinout, is combining deep machine learning techniques with artificial intelligence to study the human genome. It’s building out a database in which a user can type in a combination of mutations found in a patient – and it’ll spin out the likelihood and severity of a patient getting a disease.

“We use machine learning to try to mimic the way in which the cell works – and predict whether a person will get a disease or not,” CEO Brendan Frey said in a phone interview. And he describes it as “something akin to a Google search engine for genomics.”

This could make Deep Genomics a tantalizing new player in precision medicine.

We already have a searchable database of mutations, but what makes Deep Genomics’ approach unique is that it’s opening up a genome-wide database of more than 300 million potentially disease-causing variants, “most of which are in regions of the genome that can’t be examined using other methods,” Frey said. [...]

New company plans to revolutionize genomic medicine with deep learning

Credit: Shutterstock

Credit: Shutterstock

Featured in Gizmag, July 26, 2015.

Deep learning has already had a huge impact on computer vision and speech recognition, and it's making inroads in areas as computer-unfriendly as cooking. Now a new startup led by University of Toronto professor Brendan Frey wants to cause similar reverberations in genomic medicine. Deep Genomics plans to identify gene variants and mutations never before observed or studied and find how these link to various diseases. And through this work the company believes it can help usher in a new era of personalized medicine.

Genomic research is hard. Scientists still know relatively little about our genes and how they interrelate. But Frey and others in the field now know enough that they can equip machines to do the heavy lifting. And there's an awful lot of this heavy lifting to do. "Genomics is no longer about small datasets," Frey tells Gizmag. "It's now about very, very large datasets." [...]


Featured in Popular Science, July 23, 2015.

Tests on human genes are expensive and controversial, and no one is quite sure whether gene changes will have their desired effect. Now the Canadian startup Deep Genomics claims it has developed a computer program that can play out the different possible effects of genetic manipulation based on computers' deep learning.

Understanding how genes work is complicated because they exist in a dialogue with other genes, turning each other off or on and generating different molecules for the body to use. Researchers have been trying to understand these relationships to better treat medical conditions from cancer to schizophrenia, but the web seems to be too complicated for us to understand. That’s where deep learning comes in—using a huge dataset of people’s genetic information with its various mutations, Deep Genomics’ software can learn how cells read their genetic code and what molecules they make as a result. [...]

Toronto startup aims to shake up genome sequencing market

Featured in the Globe and Mail, July 22, 2015

A University of Toronto computer scientist known for combining artificial intelligence with big data genomics is launching a company that could create a roadmap for DNA-based therapy.

The company, called Deep Genomics, is set to launch on Wednesday.

It will be wading in to the growing market for diagnostics and personalized medicine based on whole genome squencing.

While it has become common for researchers to identify genetic mutations that appear to correlate with various diseases – thousands of mutations have been linked to cancer, for example – the technology behind Deep Genomics involves the use of computer algorithms that can tease out cause and effect relationships.

It’s a method that was developed by Brendan Frey, a professor in the university’s department of computer and electrical engineering.

The method draws on a rapidly growing discipline in computer science known as deep learning, which has lately been making inroads in a range of tough computational problems – including visual identification and speech recognition – where context plays an important role in arriving at the right answer.

Machine Intelligence Cracks Genetic Controls

Featured in, December 29, 2014

EVERY CELL IN your body reads the same genome, the DNA-encoded instruction set that builds proteins. But your cells couldn’t be more different. Neurons send electrical messages, liver cells break down chemicals, muscle cells move the body. How do cells employ the same basic set of genetic instructions to carry out their own specialized tasks? The answer lies in a complex, multilayered system that controls how proteins are made.

Most genetic research to date has focused on just 1 percent of the genome—the areas that code for proteins. But new research, published Dec. 18 in Science, provides an initial map for the sections of the genome that orchestrate this protein-building process. “It’s one thing to have the book—the big question is how you read the book,” said Brendan Frey, a computational biologist at the University of Toronto who led the new research.

'Deep learning' computers cast new light on DNA

Canadian researchers' approach may answer stubborn questions about autism and some cancers

Featured in, December 18, 2014

Researchers at the University of Toronto have developed a new way to read the human genome, which could answer stubborn questions about how flaws in DNA lead to disease.

The team, led by Prof. Brendan Frey, has used "deep learning" computer technology to read the three billion characters that represent the genome, which was first sequenced in 2003. The computers then determine which proteins, the building blocks of cells, will be produced by that DNA. 

It is thought to be the first application of deep learning to genetics. The study, by the almost entirely Canadian team of researchers, appeared Thursday in the journal Science. 

Unlocking DNA secrets with a Canadian genome search engine

Featured in Globe and Mail, December 18, 2014


Often called the book of life, human DNA has been no easy read for scientists, who face the staggering challenge of figuring out which genetic mutations lead to disease. People carry millions of them in their code, and there has been no efficient way to tell the ones that cause diseases such as cancer from those that simply make ear wax moist.

Now, a research team led by computer engineers at the University of Toronto says it has developed a biological browser, a first-of-its-kind filtering technology that may finally solve the problem.

Like a powerful search engine that mines the web for answers, the new computational system combs the human genome to seek and sort meaningful mutations. Google Inc., along with other companies, has already expressed an interest in it – raising questions about what could, or should, happen with publicly funded technology that is likely to be in demand in a growing world of Big Data.

The Dark Corners of Our DNA Hold Clues about Disease

Featured, December 18, 2014

The so-called “streetlight effect” has often fettered scientists who study complex hereditary diseases. The term refers to an old joke about a drunk searching for his lost keys under a streetlight. A cop asks, "Are you sure this is where you lost them?" The drunk says, "No, I lost them in the park, but the light is better here."

For researchers who study the genetic roots of human diseases, most of the light has shone down on the 2 percent of the human genome that includes protein-coding DNA sequences.  The trouble is, many disease-related mutations also happen in noncoding regions of the genome—the parts that do not directly make proteins but that still regulate how genes behave. Scientists have long been aware of how valuable it would be to analyze the other 98 percent but there has not been a practical way to do it.

Now Frey has developed a “deep-learning” machine algorithm that effectively shines a light on the entire genome. 


Machine learning reveals unexpected genetic roots of cancers, autism and other disorders

Featured in U of T Engineering News, December 18, 2014

"In the decade since the genome was sequenced in 2003, scientists, engineers and doctors have struggled to answer an all-consuming question: Which DNA mutations cause disease?

A new computational technique developed at the University of Toronto may now be able to tell us."