Protonation of molecules

[bricehoff]

Hello,

Thank you for creating and sharing Boltz!

Would protonating the input molecules prior to cofolding improve the accuracy of Boltz2 predictions, or is the model largely insensitive to the protonation state?

Thanks

[xiaolinpan]

I have a similar question. Given that tautomeric and protonation states are crucial for the accuracy of traditional docking programs, I am wondering if they also influence the results from Boltz.

[DrrDom]

My experiments showed that the outputs are not sensitive to tautomeric and protonation states, the poses are almost identical and predicted affinity values are very similar. It may happen that input structures are standardized in some way before prediction, because output structures of ligands in pdb are identical irrespective to the input protonation/tautomeric from (e.g. carboxylic groups are deprotonated, there were no hydrogen on either nitrogen of a quinozalinone moiety, etc). It seems that all hydrogens, polar and non-polar, are removed before predictions and thus ignored. In general this can a problem, because output structures may be unrealistic from an H-bonding point of view. Any input form can be assigned to an output structure.