A combined systems biology and machine learning approach to study patient response to anti-PD1 immunotherapy

eSMB2020 eSMB2020 Follow 2:30 - 3:30pm EDT, Monday - Wednesday
Share this

Michelle Przedborski

University of Waterloo
"A combined systems biology and machine learning approach to study patient response to anti-PD1 immunotherapy"
Anti-PD-1 immunotherapy has produced the highest response rate of any single-agent immunotherapy and has recently shown promise for the treatment of several aggressive cancers including melanoma, non-small-cell lung cancer, bladder, and head and neck cancers. However, there is high variability and unpredictability in the treatment outcome. While it remains an intensive area of research, this variability is thought to be driven by patient-specific biology, particularly, the interactions of the patient’s immune system with the tumor. Here I will introduce an integrative experimental and theoretical approach which was developed to study the patient-specific interactions between immune cells and tumor cells, to capture the variability in patient response to anti-PD-1 immunotherapy. This integrative approach utilizes clinical data from an ex vivo human tumor system that incorporates fragments from tumor biopsies in co-culture with patient-matched peripheral immune cells and plasma. The patient-derived cytokine expression levels and immune cell populations under control and Nivolumab treatment conditions were used to develop and calibrate a multi-scale systems biology model of the immune system which includes interactions of immune cells with the tumor cells and cytokine signaling. I will illustrate how the patient data was integrated into the model to capture the variability in patient response to treatment. Then I will show how the application of machine learning approaches to a simulated patient data set obtained from the calibrated model can be used to stratify features of response from non-response to anti-PD-1 immunotherapy. Next, I will discuss how transfer learning can be implemented using simulated clinical data with a subset of identified response features to significantly improve the response prediction accuracy on the ex vivo patient data. This approach has the potential to identify targeted experiments for patient screening as well as novel therapeutic targets that may sensitize otherwise non-responsive patients to anti-PD-1 immunotherapy. To illustrate this point further, I determine the optimal timing of triple combination therapy using IL-6 inhibition and recombinant IL-12 along with anti-PD1 immunotherapy to significantly improve patient response to treatment. Finally, I identify additional features of response beyond those encompassed in the systems biology model, which, while seemingly counter-intuitive, agree with recent clinical findings that may reshape our approach to cancer immunotherapy.
Hosted by eSMB2020 Follow
Virtual conference of the Society for Mathematical Biology, 2020.