< Back to Article

Which Genetics Variants in DNase-Seq Footprints Are More Likely to Alter Binding?

Fig 2

Determining which genetic variants affect TF binding.

(A) ASH p-value densities for heterozygous SNPs in different categories (the dotted blue line represents the null distribution). Numbers shown are the estimated proportion of true signal, i.e., . (B & C) Precision versus Recall operating curve (PROC) comparing CENTIPEDE predictions to (B) dsQTLs (Degner et al., 2012) [23] and (C) CTCF binding QTLs (Ding et al., 2014) [24]. For our annotation (in purple), the line is drawn for different threshold on what is considered an effect-SNP, with the (x) indicating all footprint-SNPs and the (+) indicating the default threshold of 20x difference between alleles. (B) Except for CATO (Maurano et al., 2015; dark blue) [17] and our annotation, the other prediction methods were already included in Lee et al. (2015) [18]. Note, the curve of some methods do not end at the lower-right corner because not all the dsQTLs have an annotation (e.g., if they are not in footprints). (C) For both CATO and effect-SNPs we only considered CTCF motifs, while for the methods that are not TF-centric all the scores are used. (D) Comparison of predicted binding effect for CTCF footprint-SNPs to CTCF-QTLs. Each dot represents a SNP within a CTCF binding region (ChIP-seq peak) and in a CENTIPEDE footprint with the same color annotation as in (A), the x-axis shows the predicted change in binding and the y-axis the QTL effect size for the alternate allele.

doi: https://doi.org/10.1371/journal.pgen.1005875.g002