Figure 4

Performance of peak calling algorithms at different sequencing depth. (A) Pseudo-replicates with different total mapped reads (TMRs) were created from a CTCF dataset of 36,383,621 reads. Mapped reads were twice randomly sampled to obtain fractions of 5 to 80% of the original dataset, as indicated in the panels. IDR assays comparing the pseudo-replicate datasets were performed for the predictions of the four peak-calling algorithms. (B) The number of reproducible peaks identified for an IDR threshold of 10% (IDR < 0.1) is illustrated relative to the sequencing depth. (C) Motif analysis performed with the reproducible binding sites specific to MeDiChISeq (when compared with MACS) corresponding to 29,106,897 TMRs. As illustrated, more than 40% of these sites harbor a CTCF motif (top panel; Jaspar database comparison performed by CentriMO; p-value 4.4×10^–1085) in the center of the predicted peaks (bottom panel).