Table 2 Success rate of binding site prediction of IF-SitePred and commonly used existing methods. IF-SitePred, P2Rank and DeepPocket are competive across PDB and AF2 structures, whereas FPocket experiences a significant loss of performance on AF2 structures. Success rates of top-1, top-2 and top-3 binding site prediction as measured using DCA is shown, where success is defined as the centre of the predicted binding site being within 4Å of any ligand heavy atom. We show results for IF-SitePred, FPocket, P2Rank and DeepPocket on two test sets that contain PDB and AF2 structures respectively

From: Learnt representations of proteins can be used for accurate prediction of small molecule binding sites on experimentally determined and predicted protein structures