Table 1.
Summary of the two benchmark datasets.
Figure 1.
The number of differential sites found by different methods as p-value cutoff varies from 1E-2 to 1E-8 on the ENCODE H3K4me3 ChIP-seq data.
NB = negative binomial test. GT = G-test. X-axis represents the different p-value cutoffs. Y-axis represents the number of differential sites. (A) Sensitivity curves based on the H3K4me3 ChIP data; (B) Specificity curves based on the DNA input mock data; (C) Venn diagrams of NB vs. GT using diffReps based on its default cutoff (p<1E-4).
Figure 2.
Overlap of the top differential site lists of three methods: diffReps (negative binomial test), DESeq and ChIPDiff on H3K4me3 comparing K562 and hESC.
(A) Top 2,500; (B) Top 5,000.
Figure 3.
Venn diagrams of the differential sites of diffReps (negative binomial test), ChIPDiff and DESeq on H3K4me3 comparing K562 and hESC.
Because a differential site from one method may overlap with two or more sites from another method, a priority is set for the number to be reported at the overlapped regions: ChIPDiff>DESeq>diffReps. (A) Increased sites; (B) Decreased sites.
Figure 4.
Heatmaps of correlation between chromatin modification differential sites and differential transcriptomic events on the ENCODE data.
mRNA Up, mRNA Down: overall gene expression change. Promoters: alternative promoter usage. Splicing: alternative splicing. The darkness of the color indicates the odds ratio (i.e. observation/expectation) of enrichments. The p-values of enrichments are determined by Fisher’s exact test and are labeled in red color. (A) diffReps-specific: differential sites detected only by diffReps; (B) Overlap: differential sites detected by both diffReps and another method.
Figure 5.
Distance density plot of diffReps-specific sites that are related with alternative promoter usage and alternative splicing.
The distance is determined by first looking for the closest TSS or alternative exon and then calculating the number of basepairs between the boundaries of a site and a feature. (A) Distance to TSS; (B) Distance to exon.
Figure 6.
IGV [34] genome browser screenshots of two example genes that contain alternative promoter usage and alternative splicing events.
The top two tracks are normalized genomic coverage of H3K4me3 in K562 and hESC cell lines. They are overlaid by diffReps-specific sites shown as solid bars. The bottom track is the gene model with two representative isoforms. (A) Gene MICU1 contains alternative promoter usage between two isoforms. The inset is a magnified figure of the differential sites at TSS; (B) Gene FANCI contains alternative splicing with a variant exon being preferentially included in K562 vs. hESC.
Figure 7.
The number of differential sites found by different methods as p-value cutoff varies from 1E-2 to 1E-8 on the brain H3K9me3 ChIP-seq data.
NB = negative binomial test. GT = G-test. X-axis represents the different p-value cutoffs. Y-axis represents the number of differential sites. (A) Sensitivity curves based on the H3K9me3 ChIP data; (B) Specificity curves based on the DNA input mock data; (C) Venn diagrams of NB vs. GT using diffReps based on its default cutoff (p<1E-4).
Figure 8.
Venn diagrams of the differential sites of diffReps, DESeq and edgeR on H3K9me3 comparing cocaine- and saline-treated mouse nucleus accumbens.
The priority for number reporting is set to be: edgeR>DESeq>diffReps. (A) Increased sites; (B) Decreased sites.
Figure 9.
Genomic distribution of the H3K9me3 differential sites identified by diffReps.