Fig. 7From: Learnt representations of proteins can be used for accurate prediction of small molecule binding sites on experimentally determined and predicted protein structures Number of positive labels for each residue of 10 randomly-selected structures. IF-SitePred predicted fewer residues incorrectly as ligand-binding. The frequencies of correctly-labelled (solid colour) and incorrectly-labelled (translucent) residues across 10 randomly-selected MD structures of the same protein are shown. Non-ligand-binding residues were most likely to be incorrectly labelled as ligand-binding in 0 or 1 structure, whereas ligand-binding residues had a roughly equal likelihood of being labelled as ligand-binding up to 10 times. IF-SitePred consistently had a lower rate of labelling non-ligand-binding residues as ligand-binding, with only a small reduction in the number of correctly-labelled ligand-binding residues compared to other methodsBack to article page