Structural Analysis
We used Fpocket33to identify the cavities in the available conformers (see Supplementary Table 2) with known structures of all proteins in the dataset. We estimated the pKa values ​​of the ionizable residues of these conformers using PROPKA34, to identify relevant amino acids for biological function or promiscuous activity. The residues predicted as being under positive selection were mapped to the cavities and assigned a pKa when possible. Preparation and analysis of data were done using custom scripts in the Python programming language.