The paper says: "Before the single point DFT calculations, we froze the entire structure except for the amino acid side chain of interest and re-optimized using AM1. This is a prerequisite for using the COSMO-RS parameterization BP_SVP_C21_0111, for which only the AM1 semi-empirical method gives reliable results."

My question is this: With all the known issues with AM1, do all parametrizations of COSMO-RS require an AM1 re-optimization?