Machine Learning Helps Target Cancer Patients for Immunotherapy
Approach Could Reduce Dangerous, Costly Treatment Failures
Immunotherapies are one of the most hopeful developments in the fight against cancer, leveraging people’s own immune systems to conquer previously inoperable, advanced disease. Stunning successes have propelled some of these drugs to become standard treatments, even though most of them only work for a minority of patients—sometimes less than 20 percent—while carrying significant side effects and high price tags.
Mark (Max) Leiserson, an assistant professor of computer science at the University of Maryland, and colleagues from Microsoft Research and Memorial Sloan Kettering Cancer Center are developing a new approach that uses a branch of artificial intelligence known as machine learning to cast a narrower net, better targeting immunotherapy treatment to those who will benefit.
In a study published recently in the journal PLOS One, Leiserson and his colleagues used data from a clinical trial of bladder cancer patients to show that their method could identify a suite of features that accurately predicted a key immune system response to treatment while reducing overtreatment by half.
“If your goal is to treat everyone in that particular dataset who will respond, the type of multifactorial modeling we show in this paper will let you do that while treating many fewer people who won’t respond,” said Leiserson, the paper’s lead author. Leiserson began conducting this study while he was a postdoctoral researcher at Microsoft Research, New England and continues to consult for the company.
The standard approach for the drug used in the trial relies on only two key biomarkers—features of a patient’s disease that correspond to success or failure of the treatment. That approach casts an overly wide net that includes 77 percent of the patients who wouldn't benefit. In contrast, Leiserson and colleagues showed that their multifactorial computer model predictions could cut that figure to 38 percent and still capture all the patients for whom the drug would work.
“People are realizing that predicting response is more and more appropriate and needed, and to be able to do this, the traditional kind of single biomarker approach isn’t always enough,” Leiserson said.
To generate their computer model, Leiserson and his team analyzed data from a clinical trial with a uniquely rich data set that captured information about tumor cells, immune cells and patient information such as demographics and medical history.
Recognizing the potential in such a multimodal data set, the researchers applied machine learning to the problem. They fed 36 features into their model and allowed the computer to identify patterns that could predict success.
The resulting algorithm identified 20 features that when analyzed together, explained 79 percent of the variation in patient immune responses. According to Leiserson, that percentage dropped dramatically if any one of the three categories of information was removed, indicating the multifactorial approach was critical.
The model the scientists developed isn’t ready to be used as a diagnostic tool because it incorporated data from only 21 patients, but Leiserson said they hope to improve it with more data.
“One of the goals of this work was to ask the question, ‘Should hospitals prioritize gathering this type of data?’” Leiserson said. “And now we can say that this multifactorial approach lets us better predict the response to these immunotherapies. I hope that it motivates the effort and expenditure of continuing to collect this data.”