Kerry Goetz - kerry.goetz@nih.gov
Gabriel Goodney
Regina Umarova - regina.umarova@nih.gov
Dongjing Wu
Guanghui Yang - guanghui.yang@nih.gov
"Human body epigenome maps reveal noncanonical DNA methylation variation," Nature. https://doi.org/10.1038/nature14465
This was a very complex paper with a lot of different methods and outputs. The first step was to review the supplemental material and outline the methods for each section. Check out the outline here:
Our group chose to focus on reproducing a few figures from the manuscript.
CG DMR Dendrogram
Description from supplement: "To create the dendrogram shown in Fig. 1c, we first used the cmdscale command from R to perform multidimensional scaling and compute the first 15 principal components of the CG DMR methylation level matrix. The percent variance explained from this multidimensional scaling is presented in Extended Data Fig. 1c. Next, we used the heatmap.2 function in the R package gplots18 with the default distance metric, and the Ward hierarchical clustering method on these principal components to generate the dendrogram."
Differentially Expressed Genes Dendrogram
Description from supplement: "To create the dendrogram shown in Fig. 1d, we first used the cmdscale command from R to perform multidimensional scaling and compute the first 15 principal components of the RPKM values, which were first normalized by the maximum expression value observed at each locus, from all differentially expressed genes. The percent variance explained from this multidimensional scaling is presented in Extended Data Fig. 1d. Next, we used the heatmap.2 function in the R package gplots18 with the default distance metric, and the Ward hierarchical clustering method on these principal components to generate the dendrogram."
DMR GO Enrichment
Description from supplement: "We used GREAT23 with default parameters to find functional terms of genes near CG DMRs as these terms indicate the potential regulatory functions of these CG DMRs. Since too many DMRs can saturate the Hypergeometric Test it uses, we considered at most the top 5,000 DMRs sample-specific DMRs ranked (largest to smallest) by the difference (which has to be greater or equal to 0.1) in methylation level between the hypermethylated and hypomethylated groups as input. Furthermore, we require each of these DMRs to have at least 4 DMSs. We focused on the GO Biological Process and Mouse Phenotype categories and representative results from this analysis are shown in Extended Data Fig 1e and f. The complete results are in Supplementary Tables 2 and 3."
