Deep Genomics is a Canadian start-up that combines AI and genomics to help researchers develop therapies that target all kinds of genetic-based disorders. AI is at the heart of the company’s mission, as they believe it has the potential to dramatically speed up the drug development process and create better solutions.
Deep Genomics was founded by Brendan Frey, a University of Toronto professor and one of the biggest names in Canadian AI. Back in 2002, Frey and his pregnant wife were faced with a tragic situation: they were informed by their doctor that the fetus had a genetic issue, but the limited knowledge available at the time meant it wasn’t possible to say whether it would be catastrophic or harmless.
In the end, the couple decided to terminate the pregnancy, and the experience inspired Frey to use his field of expertise to change things for the better.
He realized he could apply deep learning, a subfield of machine learning that uses algorithms that mimic the network of the brain (called artificial neural networks), to identify patterns in genetic data. With this knowledge, researchers could then develop new therapies that could potentially save lives.
“I’d come to appreciate the power of deep learning and how these neural networks could be trained to understand the data,” said Frey to the Globe and Mail in 2017. “I thought I should really use that to change society.”
The shift in focus led him to establish Deep Genomics in 2015. The group is based out of the MaRS Discovery District in Toronto and has received $56.7 million in funding to date.
Their work has already shown promise with medical riddles like Wilson’s disease through a new lead drug candidate that has been heralded as the “first ever AI-discovered therapeutic candidate.” The condition is a rare and potentially fatal genetic disease that involves excess copper storage in parts of the body.
AI drug discovery platform
Drug development is a long and expensive process, typically taking 8.5 years and costing $3.4 billion CAD. When researchers identify a protein involved in a disease, they must go through a trial and error process to find out what drug molecules in their repository could be a match for the protein’s shape. Thousands of tests may have to be conducted before a match is found, and most drug trials end in failure.
This is where Deep Genomics comes in. The team uses their AI Workbench to look at genes that contain information about these proteins and the instructions on how and when they are developed.
“There are many ways a protein could be causing a problem, resulting from different changes to the genome. We can see those changes at the level of individual genes,” said Frey to UofT News in 2017. “Instead of focusing on proteins, we’re focusing on the genetic mutations that are the source of the problem.”
Deep Genomics’ Saturn platform can examine over 69 billion different molecules covering a diverse range of tissues and diseases and identify candidates capable of manipulating cell biology.
From here, the team creates and tests 1,000 compounds to see what effect they will have on cell biology in the hope of creating new avenues for treatment. After two years, a final shortlist of three then proceed into clinical trials through collaboration with pharmaceutical companies.
“Because of the quality of their science and engineering team and the deep integration of their AI technology into their preclinical drug development pipeline, we are confident that a very large potential exists here,” said Vinod Khosla from investment partner Khosla Ventures to the Globe and Mail.
Wilson’s disease breakthrough
Wilson’s disease leads to excess copper storage in several body tissues, especially the liver, brain, and corneas. Patients with Wilson’s disease have a genetic mutation that interferes with their body’s ability to clear copper from their system.
Around 1 in 30,000 people worldwide have the condition, which can lead to liver disease, nerve dysfunction, and even death. Figuring out the genetic mutation that causes the disease has been a conundrum for decades.
Deep Genomics’ AI analyzed millions of data points and honed in on the mutation Met645Arg and the mechanism that triggers the disease, offering a clear therapeutic target. The team found that the mutation interferes with the functionality of a copper-binding protein known as ATP7B.
The software is trained to search for potential gene splicing problems from sequence data, and the alarm was raised over Met645Arg because it induces exon-skipping. The hypothesis was then validated via tests in cells where it was illustrated that 60% or more of the protein was impaired by this issue alone.
Following the creation of a shortlist of drugs designed to correct this exon-skipping mechanism, DG12P1 was announced as the lead drug candidate, and it is due to advance into clinical trials in early 2021. Frey commented that the discovery was an “important milestone” for patients and the drug discovery community.
“Within 18 months of initiating our target discovery effort, we identified a genetic mutation that causes the disease, the chemical properties needed in a molecule to target the mutation, and a compound that warrants further investigation,” said Frey in the press release.
“We are delighted to nominate the first ever AI-discovered therapeutic candidate and are eager to move it rapidly into the clinic for the potential benefit of patients,” he added.
“The clarity that this AI platform has brought to the scientific community is astounding and the potential of a therapy that could operate at the genomic level to correct the disease process is exciting,” said Frederick Askari, director of the Wilson’s disease program at the University of Michigan, in the press release.
“Patients can now have hope that a therapy may be developed that will recapitulate normal gene function and make their copper problems go away.”
Personalized medicine and the future of AI
Frey’s project has already delivered exciting results, but advances to date are only the tip of the iceberg. Fusing AI and genomics holds enormous potential for personalized medicine, according to Frey, because this technology can evaluate and interpret biology on a scale that simply isn’t possible for humans.
Deep learning’s ability to delve into vast amounts of genetic data and identify patterns among millions of medical profiles may one day help doctors to customize therapies for patients like never before. In the future, a doctor could take a patient’s genetic test results and rapidly match them up with a drug candidate tailored to their genetic makeup.
“There are some knobs that need to be turned, but I think it will happen. That’s how it’ll look,” said Frey.
Outside of medicine, Frey is a co-founder of the Vector Institute, an organization dedicated to cementing Toronto as a global AI hub. The group has received over $200 million in funding and is involved in cutting-edge AI research in areas like autonomous vehicles, robotics, deep learning, and augmented reality.
“I think in the next 10 to 20 years, almost all aspects of Canadian society will be impacted by AI, from farming to medicine to education,” said Frey in conversation with UofT News.
“AI, and deep learning in particular, is the best way to interpret data and then make rational, good choices. As the amount of data grows in all areas of society, AI will play a crucial role in making that happen.”