Rapid in silico design of antibodies targeting SARS-CoV-2 using machine learning and supercomputing

Thomas Desautels; Adam Zemla; Edmond Lau; Magdalena Franco; Daniel Faissol

To estimate ddG values using Rosetta, we started from structure minimization using the "relax" procedure. For each S protein-FAB complex model, we made 10 relaxed structures from which we chose the representative model with the lowest free energy. In these structures, we calculated ddG estimates as change of stability using the "ddg_monomer" algorithm with 50 iterations and settings as described in [32] (labeled "Rosetta ddG Total Energy" in Supplementary Materials). Additional ddG estimates were calculated using the "Flex_ddG" protocol [33] , which evaluates energy changes by focusing on interfaces within RBD-FAB complexes only (labeled "Rosetta ddG Flex" in Supplementary Materials). In addition, Rosetta calculations (Total Energy and Flex) were performed for all possible single-point mutations for each of the 31 locations selected for mutation (19 amino acid mutation possibilities *31 locations = 589 total Rosetta single-point mutation calculations).