Project 1: Computational methods

PROJECT SUMMARY – PROJECT 1: COMPUTATIONAL METHODS Project 1 will develop the general computational methods that underpin our Center’s approaches to vaccine and antibody design. In Aim 1, the Baker and DiMaio labs will extend the RoseTTAFold2 network to enable truly general biomolecular structure prediction, including full-atom models of protein post-translational

modifications, large symmetric or asymmetric protein complexes, and antibody-antigen complexes. These methods will be critical for predicting the structures of viral glycoprotein antigens from our Center’s target viral families as well as their receptors, both of which form symmetric and sometimes asymmetric complexes and

feature post-translational modifications such as N-linked glycosylation. This new All-Atom RoseTTAFold will also be used as an oracle to evaluate how strongly the sequences of our Center’s designed antigens and nanoparticle vaccines encode their target structures, allowing us to select the best designs for experimental

characterization. Finally, developing the ability to accurately predict the structures of antibody-antigen complexes from sequence alone, a long-standing goal in structural biology and immunology, will be a powerful tool for analyzing vaccine-elicited antibodies and will lay the foundation for de novo antibody design. In Aim 2,

the Baker lab will extend their machine learning methods for protein design, RF Diffusion and ProteinMPNN. These methods provide two fundamental capabilities—de novo backbone generation and amino acid sequence design—that will enable our Center to design antibodies and vaccines with unprecedented scale and

precision. In Aim 3, the DiMaio lab will combine existing machine-learning methods with newly developed methods to extend the capabilities of cryo-electron microscopy. Specifically, these methods will enable accurate structure determination from samples of proteins with various conformational states as well as

moderate-resolution datasets of antigens in complex with monoclonal or polyclonal antibodies. In our Center, structures determined using these methods will guide the design of antigens stabilized in target conformations that elicit potent neutralizing antibody responses, as well as monoclonal antibodies with exceptional

neutralizing potency and protective breadth. The computational methods developed in Project 1 will provide our Center with unique tools that power our highly innovative approaches to antibody, antigen, and nanoparticle vaccine design. All of the computational methods will be provided to the ReVAMPP network and

the broader scientific community as free, open-source software to speed biological research.