Embedding mRNA Stability in Correlation Analysis of Time-Series Gene Expression Data
Figure 3
Predicting co-regulation: ROC curves.
The ROC curves for co-regulation prediction were computed using 10 transcription factors involved in the cell cycle and assuming as true targets the DNA binding data provided by MacIsaac et al. [15], with p-value for binding 0.001. Two varying thresholds have been used for constructing the ROC curves; the higher percentile p (ranging from 50% to 90%) for the prediction of co-regulation, and the symmetric (100−p) percentile (ranging from 50% to 10%) for no co-regulation. Panel A shows ROC curves corresponding to R2 values averaged over all the 7 available datasets. Numbers below marks (circles and crosses) indicate the percentile of the distribution of R2 values used for selecting the lower and upper thresholds. Panel B shows ROC curve for each dataset obtained using as a threshold for co-regulation prediction only the 90th percentile of the corresponding distribution.