Novel immunotherapeutic drugs through computational protein design
Recent advances in computational protein design have demonstrated its usability for tasks such as designing self-assembling nanoparticles, ß-barrel membrane proteins or the rapid discovery of picomolar binders for the SARS-CoV-2 receptor binding domain as antivirals. We can further leverage these tools to help design new immunotherapeutic drugs, including stabilized viral glycoproteins for rationally designed vaccines, antibodies and antibody fragments for many applications and change the binding properties of adeno-associated virus capsids for targeted gene therapies. While computational protein design has been based on biophysical and knowledge-derived energy terms in the past, new machine learning methods are emerging with new capabilities. In this study, protein design in Rosetta was combined with the prediction of post-translational modifications using artificial neural networks. We integrated these models in the Rosetta framework, allowing the access to these predictions during design. With this, it is both possible to enrich for intended post-translational modifications while altering the sequence, e.g. for the design of N-linked glycosylation, but also to decrease the occurrence of unintended modifications sites, such as deamidations of asparagine. This new method will be applied during epitope-focused immunogen design for influenza virus vaccines and for the stabilization of antibody therapeutics.