A comparison of FoxA1 ChIP-Seq and ChIP-chip revealed the peak locations to be fairly consistent with each other (Figure 3a). Not surprisingly, the majority of ChIP-Seq peaks under a FDR of 1% (65.4%) were also detected by ChIP-chip (MAT [13] cutoff at FDR <1% and fold-enrichment >2). Among the remaining 34.6% ChIP-Seq unique peaks, 1,045 (13.3%) were not tiled or only partially tiled on the arrays due to the array design. Therefore, only 21.4% of ChIP-Seq peaks are indeed specific to the sequencing platform. Furthermore, ChIP-chip targets with higher fold-enrichments are more likely to be reproducibly detected by ChIP-Seq with a higher tag count (Figure 3b). Meanwhile, although the signals of array probes at the ChIP-Seq specific peak regions are below the peak-calling cutoff, they show moderate signal enrichments that are significantly higher than the genomic background (Wilcoxon p-value <10-320; Figure 3c). Indeed, 835 out of 1,684 ChIP-Seq specific peaks could also be detected in ChIP-chip, when the less stringent FDR cutoff of 5% is used. Another reason why peaks detected by ChIP-Seq may be undetected by ChIP-chip is that
