HOMER

mergePeaks -d <maximum distance to merge> <peak file1> <peak file2> [peak file3] ... > newPeakFile.txt
 

The program will output a new peak file containing the merged peaks to stdout.  Peaks within the distance in bp specified by "-d <#>" will be reported as the average position between the peaks found within the common region.  The origin of the peaks is specified in the 7th column of the new peak file.  Alternatively you can specify "-d given" to require a specific overlap between the start and end coordinates of the peaks.  This is more useful if comparing large regions as opposed to peaks.  The program will also output the numbers for creating a venn diagram, and these can be directed to a specific file by specifying "-venn <filename>".

Merging peaks together into a single file is very useful for certain types of analysis, such as making scatter plots that compare the tag-densities between peaks from separate experiments - in this case you want to count tags at specific and common regions.  Alternatively, you may be interested in separating the peaks into common and specific sets for focused analysis.  To do this use the "-prefix <filename>" option - this will create separate files based on overlapping peaks for each set of peaks. For example:

This will create files named "mmm_pu1.peaks", "mmm_cebp.peaks", and "mmm_pu1.peaks_cebp.peaks".

The output file will contain the following columns:

The mergePeaks program will also find calculate the statistics of co-occurrence between peaks in a pairwise fashion.  If "-matrix <filename>" is specified, HOMER will calculate statistics about the pairwise overlap of peaks.  Three separate pairwise matrix files will be produced using the supplied <filename> as a prefix:

The statistics are dependent on the effective size of the genome, which can be specified using "-gsize <#>" (default: 2,000,000,000)

Sometimes you just want to know how many other peaks bind a set of reference peaks.  If "-cobound <#>", mergePeaks counts how many of the other peak files contain overlapping peaks with the peaks found in the first peak file.  It then outputs peak files named "coboundBy0.txt", "coboundBy1.txt", etc. up to the number specified.

To find peaks that are differentially enriched between two experiments, there are two basic options.  First, you could run findPeaks (info here) using the 2nd experiment as the control sample.  Alternatively, you can use getDifferentialPeaks, which will take a given list of peaks and quickly identify which peaks contain significantly more tags in the target experiment relative to the background experiment.  To use it, follow this syntax:


By default it looks for peaks that have 4-fold more tags (sequencing-depth independent) and a cumulative Poisson p-value less than 0.0001 (sequencing-depth dependent).  These parameters are adjustable with ("-F <#>", and "-P <#>").  By specifying "-same", peaks that are similar between the two tag directories will be returned instead of differential peaks.  One caveat is that it is a good idea to set the size of the region used to search for reads to be larger than the actual peaks (i.e. +100 bp relative to the peak size) to avoid problems that arise from experiments with different fragment lengths, etc.

The output file for getDifferentialPeaks looks like the following:

Columns 8-11 contain information about the differential peak detection (colunns 1-7 come from the original peak file, including the 'score' and focus 'ratio/other' columns):

Column 8: Total [normalized] reads in the target tag directory
Column 9: Total [nomralized] reads in the background directory
Column 10: Fold change (Target/Background Total reads)
Column 11: Poisson-based p-value calculation for target read total assuming background read total represents the expected number of reads.

NOTE about normalization for this step: The read counts are normalized to the total number of reads found in their tag directories (no fancy normalization is assumed for this analysis).  However, to be conservative with the Poisson calculations, the experiment with the largest number of reads is normalized such that it contains the same number of reads as the smaller experiment.  As a result, only one of the directories - either the target directory ("Total Tags", column 8) or the background directory ("Background Tags", Column 9), will have integer counts and the other is likely to contain fractional counts.