Skip to main content
Fig. 4 | Journal of Cheminformatics

Fig. 4

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

Fig. 4

Binding site prediction on structures with decreasing accuracy. IF-SitePred outperforms P2Rank and DeepPocket when considering top-3 success on low-accuracy structures. We predicted binding sites on MD structures at increasing RMSD when aligned to PDB structures. Points represent mean success rate at each RMSD interval (value to value plus 0.25Ã…) across all 21 targets. Based on these, lines of best fit are calculated and plotted (solid lines of identical colour). Standard error of the mean across all targets is represented by the shaded areas. The three methods performed similarly when considering only the top-ranked binding site (left), with P2Rank performing slightly better at low RMSD values. However, IF-SitePred achieved higher top-3 success than P2Rank and DeepPocket at all RMSD values (right)

Back to article page