Output: predicted peaks

• 17581 genomic regions were found to have at least one sample with a peak (track hub, peaks_summary.tsv).
• 180942 peaks were detected overall, across 28 samples.
• Peak presence/absence matrices: (regions x samples), (regions x groups),
• Peak raw height matrix (regions x samples), mean coverage in peak, which is not comparable between samples, but is useful for plotting the segment mean on top of the raw coverage/count data,
• Peak normalized height matrix (regions x samples), log10(mean coverage in peak/mean coverage in background on the same sample), which is comparable between samples,
• Peak log-likelihood increase matrix (regions x samples) -- useful for ranking peaks (e.g. taking top 100 in each sample) -- wider and taller peaks are more likely, and have larger values in this matrix -- technically the values are computed as: nll_1 - nll_3, where nll_k is the negative log likelihood of the model with k segments (nll_1 is the model with one segment, no peak; nll_3 is the model with a peak, three segments, one change up and one change down).

Input: labeled genomic windows

Input: genomic segmentation problems

Output: predicted group-specific peaks

These are great candidates for re-labeling.

aorta

2497 genomic regions were predicted to have a common peak in the aorta group, and no peaks in other groups.