Figure 3. Promoters of Highly Induced Genes in the GR-CLOCK Are Enriched with High Scoring Hits from the E1-E2 Model.
(A) Left two diagrams: quantile–quantile plots show that expected likelihood (EL) scores of highly induced genes (positives) are shifted upward with respect to the control (negatives). Positives correspond to the 57 (top 0.5%; upper left diagram) or 79 (top 0.7%; lower left diagram) induced genes (ranked according to fold induction; see Methods), while the negative set consists of all remaining genes. Right two diagrams: Specificity versus number of predicted genes (sensitivity) in the group of 57 (upper right diagram) or 79 (lower right diagram). The horizontal lines represent expected specificity (lowest line), 2-fold, 4-fold, and 8-fold enrichment. Importantly, the five training genes are excluded from the set of positives in all panels. The increased specificities are highly significant: in the top row, p = 1.4 × 10−7 for 10 predicted positives (chi-squared test), p = 7.5 × 10−6 for 20, and p = 1.3 × 10−3 for 30 predicted positives. The top 30 positives are marked in blue in (B).
(B) Scatter plot representing the targetness score (fold induction in log2 units; see Methods) in function of the expected log-likelihood score of the E1-E2 model in windows of ±2,500 bp around the TSSs. Genes in blue are the 30 genes (from the group of 57) with highest match to the sequence model.