Accurate and Reproducible Functional Maps in 127 Human Cell Types via 2D Genome Segmentation

The Roadmap Epigenomics consortium has published whole-genome functional annotation maps in 127 human cell types and cancer cell lines by integrating data from multiple epigenetic marks. These maps have thereby been widely used by the community for studying gene regulation in cell type specific contexts and predicting functional impacts of DNA mutations on disease. Here, we present a new map of functional elements produced by a recently published method called IDEAS on the same data set. The IDEAS method has several unique advantages and was shown to outperform existing methods, including the one used by the Roadmap Epigenomics consortium. We further introduce a simple but highly effective pipeline to greatly improve the reproducibility of functional annotation. Using five categories of independent experimental results, we extensively compared the annotation produced by IDEAS and the Roadmap Epigenomics consortium. While the overall concordance between the two maps was high, we observed many differences in the details and in the position-wise consistency of annotation across cell types. We show that the IDEAS annotation was uniformly and often substantially more accurate than the Roadmap Epigenomics result. This study therefore reports on the quality of an existing functional map in 127 human genomes and provides an alternative and better map to be used by the community. The annotation result can be visualized in the UCSC genome browser via the hub at http://bx.psu.edu/~yuzhang/Roadmap_ideas/ideas_hub.txt