Inquiry Regarding flexible Protein-Ligand Conformation Ranking with Gnina

[ldx022]

Hello!
I have a query concerning a specific aspect of protein-ligand systems. In cases where both the protein and ligand are identical, but the final conformation varies due to sampling of the protein's pocket side chains (and in some systems, even slight alterations in the backbone), I understand that Gnina's docking power currently surpass many mainstream scoring functions. My question is: Can Gnina's rescoring be utilized for ranking in such scenarios?
To elaborate, unlike the typical problem of identifying the correct ligand pose with a fixed protein structure, here the overall pose of the protein also undergoes changes. Considering Gnina's noted efficiency in flexible docking, where side chain movements are allowed, I'm curious about the scoring process in these scenarios. Does Gnina handle the scoring/ranking in the same way for both fixed and flexible protein structures? Can Gnina's scoring function effectively discriminate the correct binding mode for the same system under these conditions? If this approach is not feasible, could you possibly suggest alternative strategies or directions for addressing this issue?
I appreciate any insights you may offer.

[dkoes]

GNINA is trained on cross-docked poses and so can deal better with non-cognate receptor structures. In the GNINA paper we evaluated flexible docking performance and found it lacking except for specific circumstances (see Fig 11 in https://jcheminf.biomedcentral.com/articles/10.1186/s13321-021-00522-2).