Fig. 5 | Nature Communications

Fig. 5

From: Community assessment to advance computational prediction of cancer drug combinations in a pharmacogenomic screen

Fig. 5

Features of poorly and well predicted combinations. a Heatmap of Pearson correlation between observed and predicted synergy scores for 118 combinations across 73 teams participating in SC1A/B. Algorithms used by each team is marked in the matrix below. b Combinations of pathways targeted. Size of node is proportional to number of drugs targeting a specific pathway in the entire dataset and width of edges is proportional to the number of drug combinations within the poorly or well predicted combination subset. Red edges highlight target pathway combinations unique to either the well predicted or poorly predicted combinations. c Types of interactions between the nearest neighboring gene and the two drug targets of poorly and well predicted combinations. Boxplots show the difference in the proportion of interactions of each type for poorly and well predicted combinations (t-test). d Proportion of poorly and well predicted combinations for different network distances (minimum number of interactions in the OmniPath shortest path) between the two targets of a drug combination, difference significance estimated with chi-squared test. e Difference in average synergy for combinations categorized by the network distance between targets are tested with t-test

Back to article page