MAnorm2 for quantitatively comparing groups of ChIP-seq samples

Abstract

Eukaryotic gene transcription is regulated by a large cohort of chromatin associated proteins, and inferring their differential binding sites between cellular contexts requires a rigorous comparison of the corresponding ChIP-seq data. We present MAnorm2, a new computational tool for quantitatively comparing groups of ChIP-seq samples. MAnorm2 uses a hierarchical strategy for ChIP-seq data normalization and performs differential analysis by assessing within-group variability of ChIP-seq signals under an empirical Bayes framework. In this framework, MAnorm2 considers the abundance of differential ChIP-seq signals between groups of samples and the possibility of different within-group variability between groups. When samples in each group are biological replicates, MAnorm2 can reliably identify differential binding events even between highly similar

HyperChIP for identifying hypervariable signals across ChIP/ATAC-seq samples

“…A count matrix and an occupancy matrix have been constructed for each data set by using MAnorm2_utils (v1.0.0) [24]. Rows and columns of both matrices corresponded .…”

“…Broad merged peaks were then divided up into consecutive bins, and narrow ones were left as they were. As for normalization, we constructed a pseudo-reference profile by averaging all the samples, and invoked the MA normalization procedure implemented in MAnorm2 [24] to normalize each sample against it (see Methods). Note also that, unless otherwise stated, each hypervariable analysis in this study targeting an individual data set has separately handled proximal and distal regions.…”

“…A complete statistical model specifying the null distribution of the scaled variances is required for assessing their statistical significances. We previously developed MAnorm2 for differential ChIP-seq analysis [24]. In HyperChIP, we follow the distributional theory proposed by MAnorm2 for modeling a group of ChIP-seq samples.…”

“…A count matrix and an occupancy matrix have been constructed for each data set by using MAnorm2_utils (v1.0.0) [24]. Rows and columns of both matrices corresponded to a list of genomic regions and the related ChIP/ATAC-seq samples, respectively.…”

“…For example, sequencing count data are inherently associated with a strong dependence of signal variability on the mean signal intensity, making the ChIP/ATAC-seq signal variability of different genomic regions not directly comparable with each other. Specifically, after a logarithmic transformation, small log-counts tend to have larger variances than large log-counts [23,24]. While this mean-variability relationship has been properly accounted for in almost all the tools for calling HVGs from scRNA-seq data, several studies called hypervariable ChIP/ATAC-seq signals by ranking genomic regions based on some variability index without considering its dependence on the mean signal intensity [9,10,19,21,22].…”