I was viewing different methods for peak calling and came across two quite similar concepts but still different methods: MACS and SICER. The main difference is that SICER uses nonoverlapping sliding windows and MACS overlapping sliding windows to identify candidate regions. However, this little difference has an impact on candidate regions, namely MACS is good for identification of TF binding sites and SICER is better for identification histone modification sites. Also, reads are more enriched for TF binging sites in ChIP-Seq then for the histone sites as they might form domains.
I am trying to understand the reason for that. Surely, SICER allows gaps in the islands/clusters/peaks, what already helps to identify larger domain of histone modification sites. However, MACS merges overlapping windows (twice the size of the sheared chromatin) using dynamic parameter. Is it because of this dynamic parameter and the nature of histone modification data that makes MACS not that good for histone modification data?
Somehow, I do not see how using either a non-overlapping or overlapping sliding window might makes one method better on one kind of data and another method on other data.