Fig. 7
From: Machine intelligence-driven framework for optimized hit selection in virtual screening
The PS-driven DNNs framework performance comparison. The PS-driven DNNs framework trained with binary fingerprint dataset encodes protein–ligand interaction pattern information, and prognostic model classify unlabeled dataset following interaction pattern information. The AUC-ROC plots shows the classification performance obtained by PS-driven DNNs framework for binary fingerprint dataset for internal evaluation 81.2% (a), for the small independent validation dataset 88.4% (b) and for benchmark dataset 89.8% (c)