Machine Learning for Cancer Immunotherapy

Microsoft Research is contributing our Artificial Intelligence and Machine Learning expertise towards important research questions at the intersection of cancer and the immune system. Cancer is the second leading cause of death in the United States. (opens in new tab) Many of us have a friend or loved one who has battled cancer, motivating us to contribute our research talents to improve cancer treatment. At the same time, drug advancements that work to harness the immune system to fight cancer are producing complex, high-dimensional datasets that can benefit from interdisciplinary attention, including from computer scientists.

Science (opens in new tab) has called cancer immunotherapy a revolution due to encouraging clinical responses from the rapid pace of drug development and FDA approvals.  We are working on this important topic in partnership with recipients of Stand Up to Cancer’s Convergence 2.0 grants (opens in new tab)and others.  These partnerships aim to improve scientific understanding of when and why immunotherapies are most likely to work. Ultimately, our goal is to help medical practitioners figure out how to most effectively target cancer immunotherapies.

Stay tuned to this space for updates as these partnerships ramp up and learn more about our involvement with other related projects below.

Improving Predictions of How Patients Will Respond to Immunotherapy

Our first immunotherapy project tested a multi-factorial machine learning model to predict patient response to checkpoint inhibitor drugs, based on a published bladder cancer dataset (opens in new tab). We integrated 19 pre-treatment tumor, immune system and clinical measurements to test how well we could predict the number of expanded Tumor Infiltrating Lymphocyte (TIL) clones in each patient’s blood 3 weeks after treatment. We found that these predictions better matched actual patient responses than predictions based on individual features, several of which are commonly used as predictive biomarkers today. As a next step, we seek to evaluate this method on larger patient cohorts.

Read the paper here: A Multifactorial Model of T Cell Expansion and Durable Clinical Benefit in Response to a PD-L1 Inhibitor (opens in new tab)

Read the article by The America Society of Clinical Oncology Post here: Machine Learning Identifies Multiple Underlying Factors Prediction Response to Immunotherapy (opens in new tab)

View the codebase here: https://github.com/lrgr/multifactorial-immune-response (opens in new tab)