Multiscale Modeling of Influenza Neutralizing Antibody and Fc Effector Biology

Seasonable inlfuenza A and B result in up to 500,000 deaths globally each year, and influenza pandemics resulting form antigenic shifts have the potential to result in in orders of magnitude higher mortality. However, seasonable influenza vaccines are only 10-60% effective, and are unlikely to provide any protection against novel pandemic strains. To address these challenges, our long-term vision is to develop computational models that predict the sequences of antibody archetypes generated in response to an influenza protein sequence, and then predict via simulation ensembles the protection afforded by a polyclonal response — with obvious and direct application to mRNA and other vaccines. Our short-term goal for this proposal is to construct generative models of the antibody repertoire and mechanistic models to investigate how a polyclonal antibody response protects against infection. Here, our overarching hypothesis is that eliciting a diverse but coordinated humoral immune response will result in more effective and robust vaccines. To understand how a pluralistic humoral immune response, with diverse neutralizing and effector antibody mixtures, is generated and coordinated to protect against the influenza virus, computational modeling is essential. Hence, we propose to develop multiscale computational models calibrated on extensive infection and vaccination datasets. These computational models will guide vaccine development by providing the tools to generate anti-influenza antibody sequence candidates, identify effective defense mechanisms at the airway mucosa, and predict vaccination outcomes in heterogenous populations. In Aim 1, we propose to develop protein language models (PLM) to capture antibody repertoire information from BCR-seq and Ig-seq data sets as high-dimensional vectors in an embedding space, using samples from healthy, infected, and vaccinated human cohorts. By comparative analysis of the distribution of antibodies in the embedding space, we will identify sequence archetypes of the anti-influenza response, and experimentally characterize the biophysical and functional properties of synthetic archetypal antibodies. In Aim 2, we propose to develop agent-based models (ABM) to capture the spatiotemporal CODEX data sets of parenchymal cells, immune cells, antibodies, and virus from ex vivo experiments in lung explants from human donors. These models will enable in silico simulations to evaluate hypotheses for how innate and humoral mechanisms provide protection at the mucosal interface. In Aim 3, we propose to develop dynamical systems models to capture nonlinear interactions between the effector functions of a polyclonal antibody response and the impact on peak viral load and clearance time from qualified clinical immunoassays from human influenza challenge cohort data sets. The overall impact will be delivery of three complimentary antibody, agent-based, and compartmental computational platforms that together will rapidly accelerate the genesis of broadly effective seasonal and universal influenza vaccines.