Fig. 6
Leveraging the effects of genetic variation to validate transcription factor binding analysis (TBA) predictions in resting macrophages. a Comparison of the mean strength of binding (number of quantile normalized ChIP-seq tags) for ATF3 in resting thioglycollate elicited macrophages (TGEMs) isolated from C57Bl/6J and Balb/cJ versus the extent of strain-specific binding. Loci with a mutation are indicated in blue (fold change ≥2) when there is strain-specific binding and gray otherwise. b Comparison of different models for predicting strain-specific binding of each monomer as measured by the Pearson correlation of a models predictions versus the extent of strain-specific binding in resting TGEMs. Models that integrate multiple motifs deltaSVM, TBA, TBA-2Strain, are represented as diamonds. Individual motifs are indicated using round points. c Frequency of mutations in significant motifs (from TBA model, p < 1e−2.5) at strain-specific (fold change ≥2) versus non-strain-specific peaks resting TGEMs. d Schematic of TBA-2Strain model. Binding sites for a transcription factor with at least one single nucleotide polymorphism or indel (red boxes) and binding sites with no mutation (gray) are identified. Next, genetic variation is quantified as the difference in the motif scores between the sequences from the two strains and then used as input to train the TBA-2Strain model to predict the extent of strain-specific binding. Model weights from the trained model indicate whether a mutation in a motif is correlated with strain-specific binding. e Heatmap of significance values for motifs that intersected between the TBA and TBA-2Strain model for each monomer in resting TGEMs. Blue indicates motifs negatively correlated with binding and red indicates positively correlated motifs
