Genetic interpretation. Giant datasets. Deep learning. This is cancer research, beyond the microscope.
A team at the University of Toronto, led by Dr. Brendan Frey, is advancing computational cancer research by developing a “genetic interpretation engine” – a GPU-powered, deep learning method for identifying cancer-causing mutations.
Today the NVIDIA Foundation, our employee-driven philanthropy arm, awarded Frey and his team a US$200,000 grant to further that work — and help them usher in an era of personal and effective cancer care.
Compute the Cure
Cancer kills almost 600,000 people each year in the U.S. alone. It can be caused by any one of an endless variety of mutations, across many different genes. This can make it hard to identify quickly and treat in a highly targeted way.
As computers grow more powerful, scientists are delving into giant datasets and deploying computer simulations to research how cancer develops.
Part of our “Compute the Cure” initiative, the NVIDIA Foundation’s grant will help Frey’s team scale up their GPU-powered methods so they can be applied to a large number of personal genomes in clinical settings, ultimately involving hundreds of thousands of genomes.
“To make a big difference in genomic medicine, we’ve developed GPU-accelerated technologies for the computationally intensive work,” Frey said. “Now, we’re focused on the next step — to change the lives of patients stricken with cancer — by experimentally validating our technologies using data from these patients.”
Leo Lee, a senior research associate in Frey’s lab, will manage the deployment of GPU-accelerated computational tools and the development of clinical experiments to validate them. Andrew Delong, who co-developed some of tools as a postdoctoral fellow in Frey’s lab, will advise the team on tool deployment and clinical validation.
In addition to demonstrating the utility of the tools in cancer biology, the team will ensure that libraries are freely available for use by other biomedical researchers working on cancer and other genetic diseases, according to Delong.
The team’s computational approach aims to overcome some of the roadblocks in personalized medicine.
At present a patient’s cancer-causing “driver” mutations have to be identified and separated from their many benign cancer-caused “passenger” mutations.
Today, a highly trained genome diagnostician can spend hours trying to understand the impact of a single mutation, pouring through databases and research papers, often coming up empty-handed.
Deep learning can help identify driver mutations more quickly, consistently and accurately than ever before. In other words, it allows human diagnosticians to scale.
The team’s approach to predicting cancer “hot spots” can learn from new genomes going forward, and, by exposing it to new data, can be trained to find causal mutations for other diseases.
With genomics data now being collected in unprecedented quantities, the University of Toronto’s project represents a groundbreaking way to analyze and gain insights from it.
“The crucial next step is validating that it works and building the clinical bridge,” Frey said. “Then the technology can be used widely and help change the lives of people with cancer.”
For more on the team’s work, watch the TedX talk by Frey.