Cancer is a wily, complex disease. To have any chance of beating it, doctors need an arsenal of tools and medications at their disposal. Chemotherapy, radiation, surgery, anti-cancer drugs and an increasing number of immunotherapies, which rev the body's immune system to fight cancer cells, can all play a role in beating back tumors. But which therapies — or combinations of therapies — are right for any one patient?
James Zou, PhD, assistant professor of biomedical data science, is hoping he can better answer that question by using an algorithm that can predict the success of a cancer treatment based on the genomic makeup of a patient's tumor — that is, the mutations that make the tumor tick.
"Two patients can have the same type of cancer, perhaps both have lung cancer, and they can be given the same cancer treatment, but they can have extremely different outcomes," Zou said. "One drug might work really well for one person and not at all for the other. Our goal with this project is to find predictive biomarkers, in this case genomic mutations, that can help inform what types of treatments are likely to benefit a particular patient."
That goal aligns well with something Zou and others in the field call precision oncology, which promotes targeted, individualized therapies that can precisely excise or destroy a person's cancer.
It's a holy grail in cancer therapy, as researchers and doctors are far from applying the concept to all patients. In a survey of more than 600,000 cancer patients in the United States, less than 8% received a treatment that targeted the mutations of their cancer.
Now, through a paper published June 30 in Nature Medicine, Zou and a team of Stanford scientists, in collaboration with researchers from Genentech and Roche, are sharing their discoveries in the hopes that it can help doctors deliver more effective treatments to cancer patients. Graduate student Ruishan Liu is the lead author.
Pan-cancer analysis
Using large datasets of cancer patients' genomic tumor information along with their electronic health records, Zou's team identified associations between mutations in eight cancer types — including ovarian, colorectal, breast and lung — various forms of therapy, and how the patient faired post-treatment.
"This is the first time we've really had this much high-quality data to fit with our computational models," Zou said.
The goal was to identify biomarkers that could flag certain mutations that render tumors more susceptible to a given treatment or, conversely, fortify them against specific treatments.
"We want to use the algorithm to answer the questions, 'How well will this patient respond to chemotherapy? Or radiation? Or immunotherapy?'" Zou said. "That means we can start with the therapy most likely to help, and we don't waste time trying out other drugs that just won't work."
Powered by data from more than 40,000 patients, the team identified 458 mutations that seemed to be linked to treatment success. A handful were existing biomarkers — a good sign that their algorithm is working properly. Another set of mutations had been cataloged from mouse or cancer cell studies but not confirmed in clinical trials or human studies. Many of the predictive markers were new. Zou's team created a website to share and visualize these discoveries.
Most of the mutations, while promising, motivate additional study, Zou said. "These will help us make really interesting new hypotheses for biomarkers, but we need more information before doctors can reliably use these mutations to guide treatment decisions."
Zou's team plans further research on a number of the mutations to determine their role in treatment successes or failures and better understand the biology behind varying treatment outcomes. The newly identified mutations could also generate leads for researchers devising drugs to target hard-to-destroy tumors.
Photo by Zenzeta
