Sometimes, starting from scratch is the best way forward.

Nearly a year into his PhD studies, Jeff found himself stuck. He had been using machine-learning algorithms to study how cancers evolve, but he found that existing algorithms were failing as the experimental data became more complex.

He had set out to map how individual cancers change over time, such as between diagnosis and relapse, or between the patient’s primary tumour and when it spreads. If successful, his algorithms could help us learn which cellular changes are driving cancers, which changes allow cancer cells to survive through therapy, and which cancer cells we should focus on treating.

Facing what seemed like endless roadblocks ahead, he realized that the existing algorithms wouldn’t cut it – he would have to start from scratch.

With a blank slate and a fresh perspective, Jeff worked alongside his supervisor Dr. Quaid Morris to simplify the problem. Every cancer cell has thousands of mutations – changes in its DNA – that distinguish it from normal cells. These mutations allow cancer cells to divide aggressively and invade other tissues. Rather than trying to understand the history of how these mutations occurred all at once, like existing algorithms did, Jeff and Quaid reframed the problem using pairs of mutations instead.

They designed a method to determine which mutation in each pair came before the other. Using this knowledge, they could then reconstruct the complete evolutionary history of a cancer, which could help predict how the disease is likely to respond to treatment. Ultimately, their methods could help clinicians choose treatments by staying one step ahead of the cancer’s mutations.

“As our collaborators acquire more detailed data from each cancer, we can help them build more precise depictions of how the disease developed, and how it may respond to treatment,” says Jeff. “These insights have enormous potential to help clinicians develop therapies that are personalized for each patient’s disease.”