Anti-PD-1 immunotherapy has been successful at treating several aggressive cancers. However, unfavorable treatment outcome or non-response to treatment have been observed in many patients, and the mechanisms underlying the response variability remain largely unknown.
To shed light on this problem, we develop an integrative systems biology and machine learning approach to study how patient-specific biology and interactions of the patient’s immune system with the tumor lead to different response phenotypes to anti-PD-1 immunotherapy. The method is built around clinical data and it enables the determination of biomarkers of favorable patient response to treatment and the identification of potential mechanisms of drug resistance.
Using these insights, we develop systems biology informed neural networks (SBINN) to calculate patient-specific kinetic parameter values and to predict clinical outcome. The SBINN approach relies on transfer learning, wherein the neural network is first pre-trained on simulated clinical data obtained from our experimentally-calibrated systems biology model , which significantly improves the response prediction accuracy on real clinical data.
Leveraging the identified potential mechanisms of drug resistance, we develop novel drug combinations and optimize the treatment protocol for triple combination therapy consisting of IL-6 inhibition, recombinant IL-12, and anti-PD-1 immunotherapy in order to maximize patient response. We also find unexpected differences in Ki-67 protein expression levels between response and non-response phenotypes, which complements recent clinical findings. Our results reflect the importance of unbiased approaches in cancer research that may challenge our pre-conceived notions and reshape our approach to cancer immunotherapy.
The approach developed in this work is generic and can be applied to other diseases and conditions associated with immune dysfunction. Moreover, it has the potential to aid in the development of targeted experiments for cost-effective patient drug screening as well as to identify novel therapeutic targets that may sensitize otherwise non-responsive patients to anti-PD-1 immunotherapy. Most importantly, the method is crucial for interpreting and learning from small clinical data sets, particularly, when the dimensionality of the data exceeds the number of samples (patients).
Please read our paper for more details.
 Smalley, Munisha, et al. "Integrating systems biology and an ex vivo human tumor model elucidates PD-1 blockade response dynamics." Iscience 23.6 (2020): 101229.