Figure 5.

Differentially enriched motifs predict Snail's regulatory output. (A, left column) Significantly enriched motifs in activated compared with repressed cobound CRMs. The right column shows enrichment of the same motifs compared with the genome. Log₂ fold enrichment values (hypergeometric P-value ≤ 0.025). (B) Work flow of machine learning approach (SVM) to discriminate between activated and repressed CRMs and the in silico mutagenesis to pinpoint the most important motifs for experimental testing. (C) Receiver–operator characteristic [ROC] curves showing SVM performance for activated and repressed CRMs. Area under the curve (AUC) is indicated. (D) The most important motifs used by the SVM to discriminate between activated and repressed CRMs (selected discriminative features). (E) In silico mutatgenesis predicts that the Tll (Tll_MA0459-1 [AAAAGTCAAM]) and ME6 (VATTWGCAT) motifs are the most important for activated cobound CRMs, affecting 50% and 33.3% of the confidently predicted activated peaks, respectively; 8.3% of CRMs depend on the Eyg motif. See Supplemental Table S9 for motif information.