Table 3.

ETNA consistently outperforms MUNK in prediction of genetic interactions between S. cerevisiae (Sce) $\leftrightarrow$S. pombe and (Spo) and S. cerevisiae (Sce) $\leftrightarrow$H. sapiens (Hsa).^a

SL prediction task	AUPRC			AUROC
	ETNA	MUNK		ETNA	MUNK
	Sce $\leftrightarrow$Spo	Sce $\to$Spo	Sce $\leftarrow$Spo	Sce $\leftrightarrow$Spo	Sce $\to$Spo	Sce $\leftarrow$Spo
Sce + Spo $\to$Sce	0.757	0.681	0.655	0.753	0.700	0.650
Sce + Spo $\to$Spo	0.773	0.606	0.554	0.744	0.581	0.571
Sce $\to$Spo	0.762	-	-	0.741	-	-
Spo $\to$Sce	0.688	0.533	0.543	0.648	0.499	0.431
	Sce $\leftrightarrow$Hsa	Sce $\to$Hsa	Sce $\leftarrow$Hsa	Sce $\leftrightarrow$Hsa	Sce $\to$Hsa	Sce $\leftarrow$Hsa
Sce + Hsa $\to$Sce	0.781	0.675	0.652	0.774	0.683	0.677
Sce + Hsa $\to$Hsa	0.738	0.485	0.553	0.803	0.396	0.582
Sce $\to$Hsa	0.650	-	-	0.687	-	-
Hsa $\to$Sce	0.686	0.552	0.642	0.684	0.545	0.664

^aFor each SL prediction task, $A\to B$ indicates SL pairs in A were used for training to predict SL pairs in B (e.g. Sce + Spo $\to$Sce indicates that SL gene pairs from both Sce and Spo were used for training, and a set of held out SL gene pairs in Sce were used for evaluation). When SL pairs for two organisms are used for training, equal numbers of positives were used (i.e. SL pairs were subsampled for the larger of the two organisms). Notably, ETNA is able to predict genetic interactions across species well using only interactions reported in the other (Sce $\to$Spo, Spo $\to$Sce, Hsa $\to$Sce, and Sce $\to$Hsa). Each of the gold standards has balanced positives and negatives (prior = 0.5). For prediction tasks based on Sce examples, MUNK failed to converge after several days of training so we were unable to calculate prediction performance.