After more than a decade of research on the use of machine learning to detect disease-causing mutations in DNA, Brendan Frey, biomedical engineering professor at the University of Toronto, this week launched his company, Deep Genomics, to bring the technology his team developed to the public at large.
Frey likens Deep Genomics’ technology to a Google search for genetic mutations: researchers can query a DNA sequence, and the system will identify mutations and tell them what’s going to happen and and why it might cause a certain disease. The system employs deep learning, a branch of artificial intelligence research.
Deep Genomics isn’t the first company to offer mutation analysis services. But Frey says his competitors don’t provide as much context: They analyze one nucleotide at a time and look for correlations between a specific nucleotide and a disease; but correlation doesn’t necessarily mean the mutation caused the disease. Frey’s team did not train their system to predict diseases, but instead to take measurements of contents within a cell (metrics such as the concentration of a specific protein) and draw conclusions about the cellular system as a whole.
“Instead of associating a mutation with a disease, our system will learn that this mutation is creating a problem because it’s causing a decrease in the protein level, and that decrease in the protein level can lead to this disease, and that’s very valuable for the diagnostician,” explains Frey. “So now if you design a drug that targets a mutation, it’ll have an effect. It’s a really big deal in that respect.”
Currently, the company’s technology is being used by several unnamed clients, who are using it in patient diagnostics. For example, labs take DNA samples from patients being tested for cancer and runs tests to detect if mutations typically associated with cancer are present; a diagnostician then examines those genetic mutations and compares them to known instances; finally, based on that information a physician recommends a course of treatment. But sometimes that second step fails—diagnosticians encounter a new, unknown mutation. That can cause complications and delays getting the right treatment under way.
Deep Genomics’ system can help diagnosticians make sense of these mystery mutations so that treatment can proceed. “It allows the diagnostician to more quickly produce the report and gives them a whole new level of information to figure out what the problem is,” says Frey.
Frey began his research 13 years ago. At the time pursuing a PhD in electrical and computer engineering at U of T, Frey and his wife at the time learned that their unborn child showed signs of a serious genetic disorder. Frustrated by the lack of knowledge doctors could give them, Frey decided to shift the focus of his research. He kept working on the machine learning techniques he had already been studying, but applied them to genetic research.
The genetic testing market is booming, currently doubling in size each year; McKinsey estimates the market will be worth $8 billion in 2018. So far, Deep Genomics has relied on angel funding and client revenue to grow, though Frey doesn’t rule out raising more as the company refines its techniques and boosts the system’s accuracy. “Right now, it’s great, it’s like being in a sandbox where we don’t have any constraints,” says Frey. “We’re just focusing on our scientific agenda and being very careful to produce the highest quality information possible.”
- Opportunities 2015: Harness “deep learning” to spot patterns
- Why robot car manufacturers should be thinking like vaccine companies: Peter Nowak
- The 3D-printed organ is coming, and Canadian firms are leading the way
- How IBM’s Watson supercomputer invents new poutine recipes