Accurate and efficient target prediction using a potency-sensitive influence-relevance voter

Table 5 AUC performance in the simulated target-prediction experiment including external validation molecules

Method	\(1\,\upmu \hbox {M}\)	\(5\,\upmu \hbox {M}\)	\(10\,\upmu \hbox {M}\)
Average AUC
PS-IRV	0.90	0.87	0.86
SVM	0.88	0.86	0.86
RF	0.85	0.84	0.84
Median AUC
PS-IRV	0.96	0.94	0.93
SVM	0.94	0.93	0.93
RF	0.93	0.91	0.90

Models are trained using 10-fold cross-validation and tested on the external validation set
Training and test sets are augmented with 1000 and 9000 random negative molecules respectively
Here, we report both average and median AUC as we find a significant difference between the two measures. The results suggest that if we exclude a few outliers, AUC performance is consistently above 0.90 for each method. Best results are in italics

ISSN: 1758-2946