Table 15.

Each of the 1,152 strategies that caused the DIKB to perform optimally in terms of sensitivity, positive predictive value, and agreement with the validation set caused the system to make the same 34 interaction predictions that were confirmed by the validation set.

pair	$\mathit{DIKB}\frac{{\mathit{AUC}}_i}{\mathit{AUC}}$	AUCi/AUC
alprazolam* - erythromycin	(0, 1.3)	1.61
alprazolam* - itraconazole	(0, 1.3) †	2.7
alprazolam* - ketoconazole	(0, 1.3) †	3.98
alprazolam* - nefazodone	(0, 1.3)	1.98
atorvastatin* - erythromycin	> 2†	1.4
atorvastatin* - nefazodone	> 2	3-4
clarithromycin - atorvastatin* ‡	> 2	≥ 1.8, max 5.4
clarithromycin* - fluconazole	> 0	≥ 1.18, max 1.33
diltiazem - lovastatin*	> 2	≥ 3
diltiazem - beta-OH-lovastatin*	> 2	3.57
diltiazem - midazolam*	> 2	3.75
diltiazem - simvastatin*	> 2	4.8
diltiazem - triazolam*	> 2	2.83
erythromycin - simvastatin*	> 2	6.3
erythromycin - beta-OH-simvastatin*	> 2	3.9
fluconazole - fluvastatin*	> 2 †	1.83
itraconazole - atorvastatin*	> 2	2.5
itraconazole - lovastatin*	> 2	14.8
itraconazole - beta-OH-lovastatin*	> 2	8.56
ketoconazole - simvastatin*	> 2	12.55
midazolam* - clarithromycin	> 2	≥ 7
midazolam* - erythromycin	> 2	4.4
midazolam* - fluconazole	> 2	≥ 2.6
midazolam* - itraconazole	> 2	10.8
midazolam* - ketoconazole	> 2	15.9
midazolam* - nefazodone	> 2	4.6
nefazodone - simvastatin*	> 2	20
nefazodone - beta-OH-simvastatin*	> 2	20
triazolam* - clarithromycin	> 2	5.3
triazolam* - erythromycin	> 2	2.1
triazolam* - fluconazole	> 2	2.5
triazolam* - itraconazole	> 2	≥ 3.1
triazolam* - ketoconazole	> 2	≥ 9.2
triazolam* - nefazodone	> 2	≥ 3.9

Asterisks indicate the drug or drug metabolite that the system predicts will be the object of the interaction.

^†The DIKB’s magnitude estimate did not correspond with validation set data

^‡The DIKB also predicted a metabolic inhibition interaction at the $\frac{{\mathit{AUC}}_i}{\mathit{AUC}}=(0,1.3)$ level with clarithromycin as the object drug.