Table 3.

For each balancing (columns) and task (rows), we report the average AUPRCs (top table) and average AUROCs (bottom table) obtained by the two Bayesian classifiers (the average is computed over the four cell lines)

Task	Balanced	Full-balanced	Unbalanced	Wilcoxon
AUPRC
IE versus IP	0.627	0.787*	0.791*	0.251
AP versus IP	0.745	0.884*	0.901*	0.066
AE versus IE	0.660	0.885	0.814
AE versus AP	0.834	0.945	0.856
AE + AP versus else	0.671	0.882	0.824
All tasks	0.707	0.877	0.837
AUROC
IE versus IP	0.82*	0.819*	0.903	0.046
AP versus IP	0.919	0.931	0.960
AE versus IE	0.893*	0.921	0.9205*	0.052
AE versus AP	–	0.960*	0.956*	0.249
	0.952*	–	0.956*	0.035
AE + AP versus else	0.929*	0.956	0.925*	0.066
All tasks	0.903	0.917	0.933

Character * marks not statistically different pairs and, in this case, the last column reports the computed p value > 0.01. Bold text highlight the best performance, when this is statistically different from all the other values