We performed 10 ns scale molecular dynamics simulations of 6 SARS-CoV-2 main protease/alpha-ketoamide inhibitor complexes in aqueous solution, in the phase before the inhibitor covalently binds to the protease's catalytic cysteine, using a polarizable multi-scale molecular modeling approach. For each simulation, 100 Mpro/inhibitor snapshots
(about 4 800 atoms) were extracted along the last 2 ns simulation segments. They were post processed using a fully quantum mechanical O(N) approach to decompose the protease in sets of fragments from which we computed the mean local interaction energies between the inhibitors and the different pockets of the protease catalytic domain. Contrary to earlier results, our analysis shows that the protease pocket S2 to be a key anchoring site able to lock within the catalytic domain an alpha-ketoamide inhibitor even before covalent bonding to the protease catalytic cysteine occurs. To target that pocket our computations suggest to consider hydrophobic groups, like cyclo-propyl or cyclo-hexyl.