Machine intelligence-driven framework for optimized hit selection in virtual screening

Table 5 Performance Comparison of CS- and PS- modules of A-HIOT on androgen receptor (AR): the performance comparison of the CS- and PS- modules of A-HIOT for internal and independent evaluation for optimized hits/leads using androgen receptor

Algorithm	Dataset	Accuracy	Sensitivity	Specificity	AUC-ROC
CS-module
CS-driven Stacked Ensemble	Internal test set (x')^a	0.867	0.845	0.902	0.864
CS-driven Stacked Ensemble	Independent test dataset^b	0.882	0.892	0.886	0.868
PS-module
PS-driven Deep Neural Networks (DNNs/DL)	Internal test set (x')*	0.852	0.823	0.894	0.879
PS-driven Deep Neural Networks (DNNs/DL)	Independent test dataset^c	0.919	0.862	0.924	0.902

^aTotal of 303 (146 inhibitors and 157 non-inhibitors) partitioned into 7:3 classified as Internal test set (x') for both CS- and PS-modules
^b1121 (249 inhibitors and 872 non-inhibitors) classified as independent test dataset for CS-module
^c878 (126 inhibitors and 752 non-inhibitors) classified as small independent validation dataset

ISSN: 1758-2946