Fig. 2
From: Uncovering correlated variability in epigenomic datasets using the Karhunen-Loeve transform
Results of functional principal component analysis for H3K4me1/me2/me3 data in H1 hESCs for TSSs of protein-coding and non-coding RNAs. (a) Heat maps of normalized coverage for H3K4me1, H3K4me2, and H3K4me3 ChIP-seq relative to TSSs of protein-coding genes. (b) First, second, and third eigenfunctions computed for different marks and for different groups of genes (black - FPC1, red - FPC2, green - FPC3). (c) Cumulative proportion of variance explained by 50 eigenfunctions for protein-coding and non-coding RNAs. X-axis is represented in log-scale. (d) Pairwise correlations between FPC scores computed for different histone marks (Y-axis) versus corresponding correlations between underlying eigenfunctions (X-axis). Results shown for 5 functional principal components. Symbol size proportional to the ranked product of variance explained by the components, red circle used for the correlation between components no. 1