Fig. 4: Differential peak analysis using imputed epigenomic tracks.

a, b show examples of non-specific and tissue-specific peaks respectively for H3K9ac in the two chosen tissues (E025 and E052). The upper part shows the measured replicates, while the lower portions display the aggregate p-value tracks for the observations and the corresponding imputations. The aggregate tracks do not capture the information on the biological variability between samples. c A scatter plot of the Wasserstein distance between the signal in the two tissues, for each peak in the enriched peakset of E025. The x-axis displays the WS distance between observed signals, while the y-axis is between imputed signals. The imputations retrieve most of the information contained in the measurements, especially for the stronger differences between tissues. We highlighted the two points corresponding to the peaks shown in (a) and (b). d Correlation heatmap of the affinity scores for different replicates. The simulated replicates correctly retrieve the expected relationships to the measured replicates, although they show a high degree of similarity between themselves, likely an artefact of the simulation procedure. e Venn diagram representing the peaks that are detected as differentially enriched between tissues using imputed and measured signals. The imputed signal retrieves 66% of the true peaks. f Binding affinity heatmaps for the measured replicates and the imputed pseudo-replicates. Each row corresponds to one of 1609 differentially enriched peaks detected in either of the measurements and imputations. The imputed replicates display the same global block structure as the measurement replicates. Source data are provided as comma-separated-values (csv) files.