. Author manuscript; available in PMC: 2015 Apr 16.

Published in final edited form as: Nat Biotechnol. 2014 Nov 17;32(12):1213–1222. doi: 10.1038/nbt.3052

Table 1.

Summary of methods and data used by each participant

Rank	PC-index	Summary	Data type used
Similarity in compound activity leads to synergy
2^a	0.60518	Identified a set of core genes defined by statistically significant DEGs in at least one compound treatment and used these genes to estimate interaction score by calculating number of overlapping genes, taking direction of regulation into account.	GD
4^a	0.57529	Computed a Pearson correlation between gene expression profiles of two compounds using genes DEGs in at least one compound treatment.	G
7^a	0.56219	Used support vector machine, trained using chemical properties, chemogenomic profiling and gene-expression data of a set of synergistic fungicidal compounds¹³.	GA
8^a	0.5507	Designed a scoring function that combines target and transporter information of each compound, DEGs, their t-score and the number of common DEGs between two compounds^38,39.	GDA
9^a	0.5327	Used the rank-aggregation method to combine results obtained from compound-pair similarity using correlation, common compound affected pathways, set of common compound-gene interactions (from ChEMBL), compound-genes interaction for one compound that are significantly affected by other compounds and compound pairs in the same clinical trial^40,41.	GPDA
11^a	0.52848	Determined cell viability by predicting activation of biological pathways in response to a single compound treatment and combined this with dose-response curves⁴².	GPD
12^a	0.52779	Used score combining overlap of gene expression signatures of individual compound treatments and cell line-specific signature derived from external datasets, taking direction of regulation into account.	GA
14^a	0.51854	Constructed probable pathways connecting compound targets and DEGs and used the Jaccard score based on gene co-occurrences in these pathways.	GPA
15^a	0.51624	Used weighted Euclidean distance, weighted by activity of each compound.	GD
31^a	0.41993	Computed correlation between gene expression profiles of two compounds using genes DEGs in at least one compound treatment.	G
Dissimilarity in compound activity leads to synergy
5^a	0.56637	Computed the Manhattan distance between pathways significantly enriched by each compound.	GP
6^a	0.56495	Designed a geometric-based score using the number of significant DEGs, the number of common DEGs between two compounds, the correlation between their gene expression profile and the dose-response curve.	GD
18^a	0.50653	Applied the Pareto ranking strategy using compound activity as well as chemical and target similarity⁴³.	GA
19^a	0.50501	Built a model that measures the effect on each of the 15 core signaling pathways by considering the number of significant DEGs, the number of common DEGs between two compounds and the direction of regulation.	GP
20^a	0.49602	Built a cooperative score by combining the number of significant DEGs, the number of common DEGs between two compounds and the correlation between weighted gene expression profiles.	G
24^a	0.46791	Identified a set of core genes defined by statistically significant DEGs in at least one compound treatment and used these genes to estimate interaction score by calculating the number of overlapping genes, taking direction of regulation into account.	G
25^a	0.45415	Estimated deviation between correlation using gene expression profile and correlation using GO terms enriched by two compounds and used that as a measure of synergy.	GP
26^a	0.44467	Built a model combining the IC₂₀ concentration of two compounds and the correlation between their gene expression profiles.	G
Combination of similarity and dissimilarity in compound activity leads to synergy
1^a	0.61303	Drug Induced Genomic Residual Effect (DIGRE) model (see main text).	GPDA
3^a	0.59981	Drug Induced Genomic Residual Effect (DIGRE) model (see main text; different cut-off for feature selection).	GPDA
21^a	0.48988	Estimated the similarity between compound pairs using DEG’s and pathway information and combined this similarity with dose-response curves.	GPDA
29^a	0.42992	Linear interpolation between two dose points on dose-response curve using similarity between each compound pair, calculated by overlap of DEGs.	GPDA
Complex synergistic relationship
10^a	0.52974	Used expression of genes, identified from the public dataset whose expression are correlated with overall survival, to predict cell viability.	GDA
13^a	0.51952	Used OCI-LY3 virtual baseline created from The Cellworks proprietary Tumor Cell Technology, trained using the known mode of actions of the compounds⁴⁴.	GPDA
16^a	0.50927	Built a model linking gene expression and cell viability. Used predicted gene expression profile after a compound combination in this model to infer the cell viability of compound pairs.	GD
17	0.50703	Identified potential effective targets by comparing expression profiles of effective and ineffective compounds and computed the sum of log-odd ratio for each compound pair under the naïve Bayes assumption.	G
22^a	0.48568	Built a bagged regression trees model using features obtained from known synergistic and antagonistic compound pairs from published literature⁴⁵.	NA
23^a	0.47183	Used expression of genes, identified from the public dataset whose expression are correlated with overall survival, to predict cell viability.	GDA
27^a	0.44346	Used the Bayesian estimation of temporal regulation and the nearest template prediction algorithm with cosine distance to associate significantly DEGs between pairs of drugs⁴⁶.	GD
28^a	0.43479	Predicted expression profile after the treatment with 2 compounds using ANOVA based liner regression and built a model linking gene expression and cell viability.	GD
30	0.42297	Used model trained using target and chemical structure of known compound combinations in cancer therapy along with protein-protein interactions.	GPA

All methods are categorized into four groups based on distinct hypotheses used by various teams. G, gene expression profile; A, additional information not provided in the challenge; D, dose-response curve; P, pathway.

Detailed method description is available in the Supplementary Note 2. The summary reported in this table was obtained directly from the participants.