An artificial intelligence system called Revolver is revealing previously hidden but common tricks by which cancers evolve to spread and defy treatment. It should allow doctors to better identify what stage cancers have reached, what they will do next and how to stop them.
We can treat cancer if we intervene before it’s too late, says Andrea Sottoriva of the Institute of Cancer Research in London, and head of the study team developing Revolver. “The key is, can you stay one step ahead of the disease?”
Revolver helped Sottoriva’s team unmask key evolutionary steps in cancers. It uses data from multiple patients to create a genetic “family tree” that tracks how cancer evolves, and identifies the series of mutations that most often lead to cancer.
Previous cancer family trees have often relied on samples from individual patients. But because mutations in cancers are so random and varied—even within one person’s cancer–the important ones can be masked by harmless background mutations and missed in the analysis.
Revolver got round this by simultaneously analysing mutation data from 178 patients at once, covering 768 tumour samples and four types of cancer—bowel, lung, breast and kidney.
This made key evolutionary steps stand out better from the background benign mutations. Three key gene mutations are already known to be crucial for benign polyps in the colon to become cancerous, for example, but they’ve never been seen together in a single patient.
Despite this, Revolver identified the three mutations as the key ones when it was tested on gene profiles from 95 colorectal cancer patients. It also correctly identified key gene mutations already known to drive evolution of lung, breast and kidney cancers.
“Understanding intra-tumour evolution during cancer therapy is critical for optimising treatment,” says Robert Gatenby of the Moffitt Cancer Center in Tampa, Florida, who has pioneered a game-theory based way of treating prostate cancer according to how it evolves. “The team has put together a very sophisticated method to do this, and I look forward to its future application in the clinic.”
Journal reference: Nature Methods, DOI: 10.1038/s41592-018-0108-x
A Product Manager with expertise in pharma marketing and sales operations