How the immune system learns

The humoral immune system learns about new antigens upon infection or vaccination, generating antigen-specific protective antibody responses. The learning "algorithm" is a mutation-selection evolutionary process that represents a stochastic dynamic process driven far from equilibrium. Recent studies show that repeated antigen exposure not only boosts past immune responses, but also generates new responses that can protect against variants. This ability to generalize upon training with a limited number of "samples" arises from feedback regulation, in which antibodies produced in earlier responses shape subsequent immune evolutionary dynamics. We focus on two questions. 1] Under what selection pressures over millennia of evolution did antibody feedback mechanisms emerge as a strategy that improves long-term protection? 2] Natural feedback is constrained by the history of past responses; if other exogenous interventions (such as antibody therapy) can be used to augment vaccines or therapies, are there optimal strategies to induce broader protection against mutable pathogens? To address these questions, we combine statistical physics-based modeling and reinforcement learning (RL). RL enables development of optimal policies for intervening actions that maximize long term rewards. Results of our calculations addressing the questions will be presented along with potential experiments to test our predictions.