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

We used molecular mechanics/generalized Born solvent accessible surface area (MM/GBSA) calculations for antibody/antigen interaction free energies [19] . MM/GBSA uses fully solvated molecular dynamics for conformational sampling of the protein complex but estimates free energy by a computationally cheaper implicit solvent model (GBSA). All molecular dynamics simulations were performed with OpenMM [34] , a toolkit for molecular simulation, using the AMBER force field [35] . Positional constraints (1 kcal/mole*Ã… 2 ) were place on the backbone atoms (C, N, and CA) during heating of the system. The system was heated in 50 K increments for 100 ps at each temperature. Once the system was at 310 K, all constraints were removed and equilibration was performed for 9 ns. Ten individual 5 ns dynamics simulations were performed from the equilibrated system for conformational sampling. The last 1 ns of the dynamics in 50 ps increments from each of the simulations was used for the MM/GBSA calculation (200 total structures). The MM/GBSA calculations were performed using MMPBSA.py that is part of the AMBER suite of programs.