Machine Learning for Cancer Immunotherapy

Machine Learning for Cancer Immunotherapy

Established: January 30, 2018



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. 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 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 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

In collaboration with Dr. Alex Snyder (Merck) and Dr. Sam Funt (Memorial Sloan Kettering Cancer Center)

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. 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

Read the article by The America Society of Clinical Oncology Post here: Machine Learning Identifies Multiple Underlying Factors Prediction Response to Immunotherapy

View the codebase here:

A Universal Map of the Human Immune System

A Microsoft Healthcare NeXT project in collaboration with Adaptive Biotechnologies

Researchers in New England are also contributing to Microsoft Healthcare NeXT’s partnership with Adaptive Biotechnologies to decode the matching between T-cells and antigens that forms the basics of the human immune system. Locked within our immune system is a record of every disease-causing bacterium, virus, or other microorganism that our body has ever fought. Deciphering this record would lead to unprecedented understanding in what our immune system is doing and will enable us to more accurately diagnose disease. However, this is a massive challenge. In partnership with Adaptive Biotechnologies, Microsoft Healthcare NeXT is driving advancements in machine learning with the goal of decoding the human immune system. For more information please see the Microsoft Healthcare NeXT site.


Previous Contributors

  • Portrait of Joe Hakim

    Joe Hakim

    Research Assistant

  • Portrait of Amy Gilson

    Amy Gilson

    Research Program Manager

  • Portrait of Francesco Paolo Casale

    Francesco Paolo Casale

    Postdoctoral Researcher