Table 4.

Performance of SARTRE on gold standards from STITCH.^a

Dataset	Confidence score	Pairs per class	Metabolite feature	$\left|{\mathbf{F}}^{\mathbf{M}}\right|$	$\left|{\mathcal{\Phi }}^{\mathcal{P}}\right|$	Accuracy	AUC	F1-measure
$D_3$ (E.coli)	400	3996	Shadow price	209	1000	0.825 ± 0.003	0.889 ± 0.003	0.835 ± 0.002
			Fingerprint	128	1000	0.816 ± 0.004	0.874 ± 0.004	0.826 ± 0.003
	700	2319	Shadow price	209	1000	0.815 ± 0.006	0.881 ± 0.004	0.826 ± 0.006
			Fingerprint	128	1000	0.802 ± 0.006	0.855 ± 0.006	0.815 ± 0.005
$D_4$ (S.cerevisiae)	400	4065	Shadow price	265	864	0.835 ± 0.002	0.894 ± 0.002	0.835 ± 0.002
			Fingerprint	128	864	0.835 ± 0.003	0.890 ± 0.003	0.834 ± 0.002
	700	1879	Shadow price	265	864	0.801 ± 0.006	0.865 ± 0.004	0.803 ± 0.006
			Fingerprint	128	864	0.800 ± 0.005	0.863 ± 0.005	0.803 ± 0.005

^aThe performance of SARTRE with respect to accuracy, AUC, and F1-measure is shown for STITCH gold standards for E.coli with 3996 metabolite–protein interacting, with medium confidence score (0.4), and 3996 randomly chosen metabolite–protein pairs not in the gold standard (assumed to be non-interacting), and 2319 metabolite–protein interacting and the same number of randomly selected non-interacting metabolite–protein pairs with high confidence score (0.7); 1000 protein features, 209 shadow prices and fingerprints of size 128 for metabolites. The table also includes the performance on the STITCH gold standard for S.cerevisiae with 4065 metabolite–protein interacting and the same number of randomly selected (non-interacting) metabolite–protein pairs with medium confidence score (0.4) and 1879 metabolite–protein interacting and the same number of randomly selected (non-interacting) metabolite–protein pairs with high confidence score (0.7); 864 protein features, 265 shadow prices and fingerprint for size 128 for the metabolites.