The immune system protects the body from a host of foreign invaders. Over the last few years, therapies that leverage these defenses to fight cancer — so-called cancer immunotherapies — have yielded impressive outcomes in combating some forms of the disease. Yet as promising as these therapies are, they currently help only a small subset of patients; the majority of patients do not respond. This picture is further muddied by the treatments’ risk of severe side effects and their high cost. And yet, there are currently no reliable biomarkers that can help physicians identify patients for whom immunotherapies will be most effective.

Click here to watch Disruptive Dozen: 12 AI Technologies That Will Reinvent Care

To lend greater clarity to the clinical use of cancer immunotherapies, researchers across the world are working on multiple fronts to uncover cellular and molecular signals that can help differentiate the responders from the non-responders. For example, a team in Boston is searching for molecular patterns in tumor samples to uncover predictive biomarkers. Their work makes use of multi-dimensional methods that scan for scores of different proteins found within tumor cells as well as neighboring cells. Given the high complexity of these data, the researchers are harnessing sophisticated, machine-learning algorithms to help propel their image analyses and pinpoint molecular patterns in patients’ tumor tissue that may signal whether a specific type of immunotherapy is likely to be effective.

The ultimate goal is to develop a kind of scoring system that can help stratify patients according to their likelihood of responding to this powerful new class of cancer drugs. These treatments encompass a range of immune-based therapies, from checkpoint inhibitors, which rev up the immune system to help destroy tumor cells to cancer vaccines to so-called CAR-T cells, which use a genetically rewired version of patients’ own immune cells to unleash a more potent attack on tumors.

Researchers are working on nearer-term solutions, too. This is because some types of cancer immunotherapies have recently had their labels expanded, creating an even more urgent need for clinical tools that can help molecularly stratify patients.

A case in point: immune checkpoint inhibitors. As of May 2017, some of these drugs are now approved for the treatment of any unresectable or metastatic solid tumor associated with the genetic abnormality known as microsatellite instability. This abnormality is currently detected clinically using either tissue or gene-based analysis.

However, performing these tests on all qualifying patients presents major technical and financial hurdles. So, a research team in Cambridge, MA applied machine learning techniques to design a method for readily identifying patients whose tumors have high microsatellite instability based on genomic characteristics (such as copy number, point mutations, and insertiondeletion events). This information is already routinely collected from cancer patients at some hospitals and therefore may not always require further testing. In those cases, this new approach offers a rapid, low-cost method for screening patients’ tumors for microsatellite instability and determining whether patients are candidates for this new treatment.

For more information about Dr. Le’s research, please contact Partners HealthCare Innovation by clicking here.