Grant: Project Grant

We will build an approach combining theory, machine learning and robotic experiments (the IMMUNOtron platform), to quantify the immune response of T cells. We are especially interested in exploring, understanding and leveraging antagonistic effects of the immune response in the context of cancer immunotherapies (CAR-T cells). Antagonism is due to the internal negative feedback loops taming immune response, and is formally related to adversarial effects in machine learning, defining the field of "adversarial immunodynamics". Such adversarial effects can in particular hinder the efficiency of immunotherapies. Adversarial effects are difficult to tame in machine learning, suggesting that targeted quantitative approaches are needed to understand and control immune antagonism. Using data generated by the IMMUNOtron platform, we will first explore how regular immune cells respond to mixture of ligands, by reverse engineering the internal feedback loops of the system from the monitoring of multiple surface markers, in multiple conditions, including mixtures of antagonist and agonist ligands which is the relevant regime in tumours . We will then move first to human cell lines, where a vast library of TCR/epitopes is available, to study the spectrum of response of human cells, and quantify the magnitude of negative effects. Lastly, we will move to engineered systems (CAR-T cells with selected ligands or receptors) to study in which adversarial regime they work, to build a predictive, mathematical model of their response, with direct clinical relevance. We will leverage the theoretical and computational toolkit developed in our group, including the adaptive kinetic proofreading mathematical framework and machine learning techniques tailored for this problem (evolutionary algorithms, biologically inspired auto-encoders). Our work will allow us to precisely identify the relevant regimes for immunotherapy, with direct clinical applications.