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

For our ddG FoldX calculations of 89,263 mutant antibodies in complex with the RBD, we started with structure minimization using the "minimize structure" procedure available in UCSF Chimera [31] . This process was followed by up to 50 iterations of the "RepairPDB" function in FoldX. We performed ddG calculations on the defined list of mutations using the "BuildMODEL" procedure, which estimates stability (dg) defined by the free energy of a protein. We used up to 46 runs in this procedure to check if calculated rotamers of specified mutations converged to the optimal or trapped solution. A first set of our ddG estimates was calculated as a difference in free energy (dg) between the mutant and wild-type (labeled "FoldX ddG Average" in Supplementary Materials). Additional ddG estimates were calculated using the "AnalyseComplex" algorithm for which we used the wild-type and mutated models to calculate ddG based on energy changes in interface only within RBD-FAB complexes (labeled "FoldX ddG Interface" in Supplementary Materials).