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

In the absence of a known SARS-CoV-2 spike protein structure, we characterized the SARS-CoV-2 surface glycoprotein sequence YP_009724390.1 [13] by constructing a homology-based structural model using the AS2TS protein modeling system [14] . The structures of the spike proteins from SARS-CoV-1 (Protein Data Bank (PDB) entries: 5x58 [15] , 6nb6 [10] , 2dd8 [11] , and 3bgf [16] ) were identified as the closest and most complete PDB structural templates to use for modeling, with resolutions 3.2, 4.2, 2.3, and 3.0 Ã… respectively. Of the four templates, the last three provide experimentally solved complexes with three antibodies, S230, M396, and F26G19. As our main focus in constructed homology models was to achieve the highest possible accuracy in the RBD-FAB (antigen-binding fragment) interfaces we used additional PDB templates to refine models in corresponding regions. For example, PDB template 2ghw [17] was used to improve accuracy in the modeled interface between RBD and FAB of neutralizing antibody 80R.