Table 2.

Summaries of reviewed systems approaches for identifying drug combinations

Disease models	Method	Key findings	Validation	Reference
Computational models of cell signaling networks
Breast cancer	Mass-action model	Combined inhibition of MEK and PI3K optimally decreased cell viability.	in vitro	[25]
Ovarian cancer	Mass-action model	the ratio of PTEN to activated PI3K predicts RTK inhibitor resistance	in vitro	[26]
Ovarian cancer	Mass-action model	ErbB3 inhibition inhibits the ErbB-PI3K network more potently than current therapies.	in vivo (rodent)	[27]
Breast cancer	Logic-based	Combined inhibition of c-MYC and ERBB2 improved treatment for trastuzumab resistant breast cancer.	in vitro	[30]
T cell large granular lymphocyte leukemia	Logic-based	Sphingosine kinase 1 and NFKB are essential for survival of leukemic T cell large granular lymphocytes.	in vitro	[31]
Colorectal cancer	Fuzzy Logic	MK2 and MEK are co-regulators of ERK and EGF induced IKK inhibition.	in vitro	[32]
Cardiac hypertrophy	Normalized-Hill model	Ras had the greatest influence on hypertrophy and correlation between node degree and influence is low.	in vitro	[35]
Various	3-node enzymatic models	Identified consistent synergistic and antagonistic motifs.	in silico	[41]
Various	4-node enzymatic models	Synergy is more prevalent in motifs with negative feedback between the target and an upstream effector or mutual inhibition between parallel pathways.	in silico	[42]
Cardiac hypertrophy	Statistical association model	Maladaptive and adaptive hypertrophy features were in separate modules in the simplified hypertrophy network map generated by k-means clustering of ligands and phenotypic outputs.	in vitro	[45]
Melanoma	Statistical association model	PLK1 inhibition increases cytotoxicity of RAF inhibitor resistant melanoma cells.	in vitro	[47]
Various	Statistical association model	Reconstructed classic T cell signaling network using multiparameter single-cell data and Bayesian network inference.	in vitro	[48]
Signature-based approaches
Lung cancer	CMap	PI3K inhibition enhanced docetaxel-induced cytotoxicity	in vitro	[55]
Lymphoblastic Leukemia	CMap	mTor inhibition induced glucocorticoid sensitivity by decreasing MCL1	in vitro	[52]
Lung cancer	K-Map	The combination of bosutinib and gefitinib has synergistic effects in EGFR mutant non-small cell lung cancer	in vitro	[57]
Network-based approaches
Osteosarcoma	Target Inhibition Map (TIM)	Developed an algorithm using a training set of drug sensitivities with known targets to predict responses to new drugs and combinations.	in vitro	[58,59]
Breast and pancreatic cancer	TIMMA	Target Inhibition inference using Maximization and Minimization Averaging (TIMMA). Improved computational cost and accuracy of the above TIM approach. Predicted kinase pairs that could be inhibited to prevent cancer survival.	in vitro	[60]
Various	Elastic Net Regularization	Performed phenotypic screen using an optimal set of 32 kinase inhibitors. They used an elastic net regulatization algorithm to deconvolute the polypharmacology and identify key kinases regulating cell migration.	in vitro	[61]
Lung and breast cancer	DrugComboRanker	Created drug and disease functional networks based on genomic profiles and interactome data. Drug combinations are predicted by identifying drugs whose targets are enriched in the disease network.	Literature support	[62]
Various	Mixed integer linear programming	Built a network of drug-target interactions from DrugBank. Given an input gene set, the algorithm selects drug combinations that maximize on target effects and minimize off target effects	Literature support	[63]
Various	Systems analysis of Drug Combinations	Drugs with similar therapeutic effects cluster together in a network of successful drug combinations produced using the Drug Combination Database [59]. Network observations were used to develop a statistical approach for predicting drug combinations (DCPred)	Literature support	[65]
Drug-drug interactions	Drug-drug interaction network	Applied five machine learning models to a data set of drug-drug pair similarities including 721 approved drugs to predict drug-drug interactions.	Literature support	[66]
Integration of functional genomics and computational methods
Breast cancer	RNAi screen	PTEN downregulation with active PI3K signaling induce trastuzumab resistance	in vitro	[68]
Colorectal cancer	RNAi screen	EGFR inhibition synergizes with BRAF(V600E) inhibition	in vivo (rodent)	[69]
Lymphoma	8-gene RNAi signature	Drug combination signatures were usually a weighted composite of single drug effects	in vitro	[70]
Colorectal cancer	RNAi screen	The combination of Selumetinib (MEK1/2 inhibitor) and CsA (Wnt inhibitor) has synergistic anti-proliferative effects	in vivo (rodent)	[71]
High-throughput drug combination screens
HIV	Pooled screen	Used pools of 10 drugs in 384-well plates to study all possibly pairs of 1000 compounds in the minimum number of wells possible	in vitro	[72]
Melanoma	Drug combination screen	Sorafenib (a multi-kinase inhibitor) and diclofenac (NSAID) had synergistic effects across all nine tested melanoma cell lines.	in vitro	[73]
Lymphoma	Drug combination screen	Screen of 500 compounds with ibrutinib revealed favorable combinations with inhibitors of PI3K signaling, the Bcl2 family, and B-cell receptor pathway	in vitro	[74]
Various cancers	Drug combination screen	Screen of 5,000 combinations of FDA-approved drugs in the NCI-60 cancer cell line panel.	in vitro	[75]
Lymphoma	RNAi-modeled tumor heterogeneity	Intatumor heterogeneity influences the prediction of effective drug combinations.	in vivo (rodent)	[77,78]