Figure 1.
Workflow for the pipeline used to normalize and analyze Connectivity Map microarray experiments.
The reliability of drug-induced gene expression profile similarity scores (DIPS scores) were evaluated using independent drug features as benchmark. Using the processed data, differential regulation of drug-induced drug targets was investigated.
Figure 2.
Analysis of systematic biases and benchmarking with independent features of chemicals.
(A) Distributions of the DIPS scores for the pair-wise comparisons of gene expression profiles constructed using biological controls and mean centering as background across four drug/batch categories: i) both profiles are from the same drug and the same batch (Blue), ii) the same drug from different batches (Red), iii) different drugs from the same batch (Yellow) and iv) different drugs from different batches (Grey) (B) ROC curves are used to assess the performance of the DIPS score (blue line) and provide a comparison with the method described in Iorio et al. (red line) [9]. Area under the curve values for each ROC curve: Chemical structural similarity: AUC (DIPS = 0.028 for FPR<0.1) and (AUC Iorio et al. = 0.016 for FPR<0.1). For 4th level ATC sharing, the AUC (DIPS = 0.016 for FPR<0.1) and (AUC Iorio et al. = 0.009 for FPR<0.1)(Refer to Figure S2 for the complete ROC plots.).
Figure 3.
Drug-induced differentially regulated drug targets.
Anova is used to assess the significance of the differential expression of drug-induced drug targets against the mRNA changes of the same gene in the population of heterogeneous drug treatments from CMap. The genes are mainly ordered based on their q-values as provided in Table S1. In the scatter plots, inhibitors/activators are labeled in red/green respectively and grey represents all other treatments present in CMap.