#Removing redundant predictions and #re-ranking #increases recall by 5% for #VN-EGNN and #DeepPocket-SEG.

Removing #redundancy, #re-scoring using sum of squares of the #ligandability #score we defined and re-ranking of remaining pockets increases #IF-SitePred #recall by 13%.

#IF-SitePred does not report a residue #ligandability #score.

However, using their #LightGBM models #probabilities we define one. This LS captures ligand binding sites that are #not #reported by this method.

This method would #benefit from a residue-level score. Here 8 examples.