Algorithm comparison for HNF4 α and H3K4me3. This figure shows how the peaks called by BayesPeak, ChipSeq Peak Finder, MACS and PeakSeq compare. For CSPF we used a maximum gap of 100 bp between reads to create clusters, which were then identified as peaks if they contained more than 9 reads, more than 5 of which overlapped at the same base, and if the total count was larger than the corresponding number in the Input sample by a factor of 5. To identify the enriched regions using the other methods, we chose p-values smaller than 0.05 for both MACS and PeakSeq. The number of peaks called by BayesPeak seems smaller in these diagrams than we report in the paper because in some cases more than one region is covered by a long peak identified by another method; in this case, they are merged into a single peak covering the entire region.