Fig. 2

Evaluation of the three imputation approaches at genomic positions that show variation in signal across cell types. a A schematic describing how genomic loci are segregated on an example of four cell types. MACS2 peak calls (in gray) are summed over each of the cell type. Genomic loci are then evaluated separately based on the number of cell types in which a peak occurs. b Each panel plots a specified performance measure (y-axis) across varying sets of genomic positions (x-axis) for the H3K4me3 assay. For each point, genomic positions are selected based on the number of cell types in which a peak is called at that position, up to a maximum of 127. MSE is calculated between H3K4me3 ChIP-seq signal and the corresponding imputed signal. Precision and recall are computed by thresholding the imputations at 1.44 and comparing to MACS2 narrow peak calls on the corresponding experimental signal. In the plots, the series labeled “Roadmap” use the experimental Roadmap data likewise thresholded at 1.44 and compared to MACS2 narrow peak calls. c Similar to b, but using DNAse-seq instead of H3K4me3. All analyses are restricted to chromosome 20