Figure 7

Position Weight Matrix extension enables high fidelity identification of Predicted Binding Regions in ChIP-seq peaks. Three different PWMs were constructed for CAD (red), HB (blue), KNI (green), and KR (purple) respectively, corresponding to; the core PWM-PBR (core), a region encompassing the core PBR and all statistically significant flanking sequence context-dependent biases (extended), and the core with 25 bp of flanking sequence on each side (full). For each individual PWM, PATSER was run on filtered ChIP-seq peaks using a low stringency (A, corresponding to the 25th percentile) or high stringency (B, corresponding to the 50th percentile) primary cutoff and then a secondary cutoff score corresponding to the 0th percentile (see Methods for full details). Charts depict the true positive rate (defined as the number of ChIP-seq peaks for which the PWM predicts > 0 binding regions divided by the total number of ChIP-seq peaks). Statistical significance of divergence in true positive rate of expanded PWMs from the corresponding core type PWM is indicated with asterisks (Two-tailed Fisher exact test, * denotes P-value < 0.05, ** denotes P-value < 0.0001).