Machine Learning to Modulate Influenza Immunity

PROJECT SUMMARY/ABSTRACT In this proposal, we will leverage "Machine Learning to Modulate Influenza Immunity" by developing and implementing computational tools tailored towards the design of broadly protective vaccines against influenza viruses. Machine Learning (ML)-based algorithms pioneered by us and others have revolutionized protein structure prediction (AlphaMask, AxIEM, AlphaFold, or ESM) and design (ProteinMPNN, Rosetta) and have improved accuracy and speed of protein engineering. Here, we will integrate Artificial Intelligence/Machine Learning (AI/ML) and traditional computational biology to create a novel class of immunogen candidates that overcome the inefficacy of current influenza vaccines which fail to elicit broad and long-lasting protection. We will explore three immunogen design strategies: 1.) germline-targeting on the lateral patch of hemagglutinin (HA), one of the most conserved epitopes in the highly immunogenic HA head domain; 2.) HA heterotrimerization to create immunogens exposing various HA types and/or subtypes; and 3.) epitope-focusing to shift the immune response from highly variable regions to more conserved epitopes. Each aim will develop a computational pipeline with new algorithms for their respective immunogen design objectives that will be iteratively refined by experimental feedback, thereby successively improving the performance of the computational tools. In vitro experiments will provide high-throughput feedback from mammalian surface display, including deep mutational scanning of germline-targeting mutations and deep glycosylation scanning to evaluate the potential for immune evasion through hyper-glycosylation of HA. These large datasets will be integrated in AI/ML tools to extend predictions to other strains and further guide future vaccine design. To validate the effect of novel immunogen candidates, we will employ human Ig loci transgenic mice capable of producing fully humanized B-cell receptors. To simulate the effects of original antigenic sin, a mechanism in which primary contact with a novel immunogen has a lasting impact on subsequent antibody responses, we will analyze immunogenicity in animals pre-immunized with wildtype HA, thereby obtaining insights into the interaction of computationally designed immunogens with human-like antibody responses. The in vivo experiments will be complemented by ex vivo sorting of human naïve and memory B cells with computationally designed immunogens. All experiments will be supported by structural characterization through computations, X-ray crystallography, electron microscopy, and mass spectrometry to identify determinant mechanisms of protections. The efficacy of promising immunogens will be tested in lethal challenge experiments. We ensure the AI/ML-based computational tools developed in this project are readily available free-of-charge and transferable to other immunogen design challenges. The deepened understanding of the interaction of the immune system with rationally designed immunogens will support the future development of vaccines.