Fig. 2From: Sequence-based prediction of protein binding regions and drug–target interactionsGeneration of BRs from 3D complex databases. BRs were generated using by the following procedure. We collected BI from the scPDB and PDBBind databases. The collected BI was mapped onto the corresponding UniProt sequence, and as shown, it can be in non-consecutive sequences. To build consecutive subsequences including BI, we expanded the mapped BI to a length of 9 and merged the overlapped BI. In the BR training steps, we fed randomly generated BRs with an IoU larger than 0.7 to the prediction modelBack to article page