Fig. 4
From: Sequence-based prediction of protein binding regions and drug–target interactions
Prediction performance for drug–target interactions in the independent test datasets. DTIs were evaluated using PubChem Bioassays. Error bars stand for standard error. a DTI prediction performance for bioassays with all druggable proteins derived from DGIdb. HoTS overall outperformed the other models and HoTS without BR training, showing that BR training indeed improves DTI prediction performance. b DTI prediction performances for a subset of bioassays on DTI test dataset 1, whose proteins have the same SCOPe family as the BR training dataset. HoTS overall greatly outperformed the other models and HoTS without BR training, showing that BR training indeed helps DTI prediction with the same structural motifs