TABLE 2.

Biological assumptions, sources of input data, and network statistics for predictive models.

		Synthetic lethal interactions, biological assumptions.	Input data sources						Statistical tests
			Gene expression	SCNA	Somatic mutation	Phylogenetics	Clinical patient data	Short hairpin RNA
Statistical inference	DAISY Jerby-Arnon et al. (2014)	Gene pairs that overlap across all assumptions.	☑	☑	☑			☑	Wilcoxon rank sum, followed by Bonferroni correction for multiple hypothesis testing; gene co-expressions were calculated using Spearman correlation
		1. Survival of the fittest (SoF): Synthetic lethal pairs are co-inactivated for cell death.
		2. Death upon single gene knockdown when another gene is inactive is synthetic lethality.
		3. Synthetic lethal pairs are co-expressed.
	Srihari et al. Mutual Exclusivity Model (Srihari et al., 2015)	Gene pairs that are frequently altered in a mutually exclusive manner are defined as synthetic lethal.	☑	☑					The statistical significance value was obtained by subtracting SL score obtained by hypergeometric test from 1: $p_{val}=1-S{L}_{hypergeometric}$
	ISLE Lee et al. (2018)	Gene pairs that exhibit the following characteristics:	☑	☑	☑	☑	☑		Statistical significance tests used for the respective assumptions:
		1. Gene pairs are rarely co-inactivated compared to their individual inactivation frequencies.							1. Hypergeometric test
		2. Gene pairs yield better patient survival through their co-inactivation, reducing tumor fitness when co-inactive.							2. Likelihood ratio test
		3. Gene pairs tend to co-evolve and thus have high phylogenetic similarity.							3. No statistical test at this step
									Afterward, Wilcoxon rank sum was used to compare identified SL pairs with drug target response
	ASTER Liany et al. (2020a)	Gene pair (Genes A and B) that passes the following tests:	☑	☑	☑	☑			Wilcoxon rank sum, followed by Fisher’s method for combining significance p-values. False discovery rates were determined using the Benjamini–Hochberg method
		1. For tissue-specific samples with high Gene A copy number, the expression level of Gene A is significantly higher than that of non-cancerous samples of the same tissue type.
		2. For tissue-specific samples with high Gene A copy number, but low Gene B copy number, expression level of Gene B is significantly lower than that of non-cancerous samples of the same tissue type.
		3. Expression levels of Gene A in Test 1 is significantly higher than those of Gene B in Test 2.			☑
	SLIdR Srivatsa et al. (2019)	Synthetic lethal pairs consist of a significantly mutated gene and its interacting genes that yield cell death upon co-occurrence of their aberrations.		☑	☑			☑	Custom, rank-based statistical test was used where the p-value was obtained from the lower-tail probability
	MiSL Sinha et al. (2017)	The mutations of synthetic lethal pairs are amplified more frequently and are deleted less frequently while in concordance with their gene expression profiles.	☑	☑	☑				Fisher’s exact test for evaluating gene-pair behavior dependence, followed by two-tailed unpaired Student’s t-test
Network-based models	VIPER Alvarez et al. (2016)	A probabilistic framework where tissue-specific gene-expression data are used to identify regulator-target interactions following the activation or repression of a regulator.	☑						Analytic rank-based enrichment analysis (aREA) statistical analysis is used to discern differential gene activity
	OptiCon (Hu et al., 2019)	Using gene expression profiles in a regulatory network, optimal control nodes (OCNs) are identified such that they exert maximal control over deregulated pathways, but minimal control over unaffected pathways for a given disease. For SL tasks, OCNs point to potential synthetic lethal pairs	☑	☑					Wilcoxon rank test and one-sided Kolmogorov-Smirnov test