Evaluation of an Affinity-Enhanced Anti-SARS-CoV2 Nanobody Design Workflow Using Machine Learning and Molecular Dynamics

<i>In silico</i> optimization of protein binding has received a great deal of attention in the recent years. Since <i>in silico</i> prefiltering of strong binders is fast and cheap compared to <i>in vitro</i> library screening methods, the advent of powerful hardware and advanced machine learning algorithms has made this strategy more accessible and preferred. These advances have already impacted the global response to pandemic threats. In this study, we proposed and tested a workflow for designing nanobodies targeting the SARS-CoV-2 spike protein receptor binding domain (S-RBD) using machine learning techniques complemented by molecular dynamics simulations. We evaluated the feasibility of this workflow using a test set of 3 different nanobodies and 2 different S-RBD variants, from <i>in silico</i> design and bacterial expression to binding assays of the designed nanobody mutants. We successfully designed nanobodies that were subsequently tested against both the wild-type (Wuhan type) and the delta variant S-RBD and found 2 of them to be stronger binders compared to the wild-type nanobody. We use this case study to describe both the strengths and weaknesses of this <i>in silico</i> assisted nanobody design strategy.