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

Starting on Jan 23, 2020, we used two high-performance computers at LLNL over the course of 22 days to support over 200,000 CPU hours and 20,000 GPU hours, performing 178,856 in silico free energy calculations of candidate antibodies in complex with the SARS-CoV-2 RBD. We performed this work using just the SARS-CoV-2 sequence and previously published neutralizing antibody structures for SARS-CoV-1. Our predicted structures of SARS-CoV-2 proved to be accurate within the targeted RBD region when compared to experimentally derived structures published weeks later. With our predicted SARS-CoV-2 structures, we used our in silico design platform to evaluate 89,263 mutant antibodies by iteratively proposing mutations to SARS-CoV-1 neutralizing antibodies to optimize binding within the SARS-CoV-2 RBD. We selected 20 initial antibody sequences predicted to target the SARS-CoV-2 RBD. Starting from a baseline free energy of -48.1 kcal/mol (± 8.3), our 20 selected first round antibody structures are predicted to have improved interaction with the SARS-CoV-2 RBD with free energies ranging as low as -82.0 kcal/mole. The baseline SARS-CoV-1 antibody in complex with the SARS-CoV-1 RBD has a calculated free energy of -52.2 kcal/mole and neutralizes the virus by preventing it from binding and entering the human ACE2 receptor. These results suggest that our predicted antibody mutants may bind the SARS-CoV-2 RBD and therefore potentially neutralize the virus. Additionally, our antibody mutants score well according to multiple antibody developability metrics via the Therapeutic Antibody Profiler.