Fig. 4

Transcription factor binding analysis (TBA) identifies motifs predicted to specify differential AP-1 monomer-binding in resting thioglycollate elicited macrophages. a DNA motifs rank order based on the significance of the motif according to the likelihood ratio test. The black box represents the most significant motifs positively correlated with binding for all AP-1 monomers and are listed in (i) and the most significant motifs negatively correlated with binding for all AP-1 monomers are shown in the gray box and are listed in (ii). The significance of motifs positively correlated with binding that show a 100-fold likelihood difference between two monomers are shown on the left heatmap (red); the right heatmap (blue) gives the significance of corresponding motifs negatively correlated with binding. b Comparison of the performance of TBA against the AP-1 motif score alone, Bayesian Markov Model (BaMM) motif score, and gapped k-mer SVM as measured by the area under the receiver operating characteristic curve (auROC). Error bars indicate the standard deviation of auROC across 5 cross-validation sets. c Number of motifs that pass an in-silico mutagenesis test for significance (the likelihood ratio test comparing the performance of a full model that uses all the motifs and a mutated model with one motif removed) at various p-value thresholds. d Predictive performance of TBA when predicting ATF3, Jun, and JunD binding as motifs are iteratively removed starting from the least important motif based on the weights calculated by TBA. Inset shows performance values beginning at 150 motifs removed where predictive performance begins to drop