Figure 1.
Histone modification profile as seen in the UCSC genome browser [17].
Tracks 1 (red) and 2 (green) show the H3K27me3 modifications for ES and NP cells, respectively. Track 3 shows the differentially enriched regions found by our ChIPnorm method. Track 4 shows the differentially enriched regions found by the ChIPDiff method [9]. In tracks 3 and 4, red indicates differential enrichment in ES cells and green indicates differential enrichment in NP cells. Track 5 shows the UCSC genes.
Figure 2.
We see how we can get the ChIP-seq library, input DNA control, and the random distribution (null hypothesis).
Figure 3.
The new inverse cumulative distribution function on the modified libraries (after stage 1).
On the axis is the percentile, on the
axis are the bin values.
Figure 4.
The schematic diagram of the ChIPnorm method.
In the first stage one we find the enriched-significant bins by removing various kinds of errors in the data. In the second stage we normalize the two ChIP-seq libraries and find differentially enriched bins.
Figure 5.
Enrichment level of bins with respect to gene density in a 1 Mbp region.
The axis indicates lowest to highest gene density. (a) The
axis indicates the total number of H3K27me3 ChIP-seq fragments divided by the number of Mbp regions (counts per megabase) found in each gene density. (b) The plots are re-normalized so that the
axis range is same for both ES and NP cell data. We see that the enrichment level of ChIP-seq data increases with respect to gene density for both ES and NP cells.
Figure 6.
Enriched bins with respect to gene density in 1 Mbp region.
The plots are normalized. axis 1 to 10 indicates lowest to highest gene density, while
axis 0 to 1 indicates minimum to maximum average number of differentially enriched bins for both ES and NP cells. Blue line indicates ES differentially enriched bins and red line indicates NP differentially enriched bins. The number of enriched bins per 1 Mbp should increase with gene density.
Table 1.
Sensitivity analysis percentages using various methods (data from Mikkelsen et al. 2007 [8]).
Figure 7.
Plot of over five different threshold values.
(a) NP is differentially over-expressed compared to ES, (b) ES is differentially over-expressed compared to NP.
Figure 8.
Robustness studies: sensitivity and error analysis for ChIPnorm by fixing the fold-change threshold and varying the bin size from 200 bp to 2000 bp in steps of 200 bp.
Data from Mikkelsen et al. 2007 [8].
Table 2.
Sensitivity analysis for human ES and GM12878 cells (replicate 1 data from ENCODE Broad database) percentages using various methods.
Figure 9.
Two ROC curves are shown for various methods.
(a) first ROC: Class 1 – four-fold NP differentially over-expressed genes compared to ES; Class 0: rest of the genes. (b) second ROC: Class 1 – four-fold ES differentially over-expressed genes compared to NP; Class 0: rest of the genes.
Table 3.
False-positive rate (FPR) analysis for human ES cells (H3K27me3 data from ENCODE Broad database) for the two replicates.
Figure 10.
Gene profile according to expression and histone modifications.
Genes are grouped in (A–E) according to increasing ratio of expression level in ES cells and NP cells. Within each groups, genes are classified into 4 types. Type 2 genes have differential histone enrichment in ES cells in their promoter regions and type 4 genes have differential enrichment in NP cells. (a) Percentage of type 2 genes is decreasing, while percentage of type 4 genes is increasing along the group (A–E). (b) Percentage of type 2 genes is increasing, while percentage of type 4 genes is decreasing along the group (A–E).
Figure 11.
Bivalent regions in genes; (a) in 333 selected genes; (b) in UCSC known genes.
Genes are attributed to classes according to the presence of modifications in ES and NP cells. The “A–B” notation in the labels indicates the presence of modification of type “A” in ES cells and modification of type “B” in NP cells. (*) marked labels have bivalent domains in ES cells.
Figure 12.
Bivalent gene profile vs. expression data.
Genes are grouped in (A–E) according to increasing ratio of expression level in ES cells and NP cells. Each bar shows the percentage of genes with the corresponding “A–B” modifications (as listed in the box), “A” for modifications in ES cells and “B” for modifications in NP cells. It is seen that there is a strong over-representation of the K4+K27-K4 transition (yellow) in the genes class, which is strongly up-regulated in NP cells.