...

Using the effective fragment length calculated by Cross Strand-Correlation, each read is extended in the 3' direction by the effective fragment length and overlapping extended reads are merged into single peaks. For paired-end reads, the distance between paired reads is used as the fragment length and overlapping fragments are merged into peaks. For peak detection, the genome is divided into windows of a user-defined size and the number of fragments whose mid-points fall within each window is counted. A model for expected read density (a zero-truncated negative binomial) is used to determine which peaks are significantly enriched over a user-defined false discovery rate (FDR). See the ChIP-Seq white paper for more information on the peak-finding algorithm and tips for setting the Fragment extension and window sizes. 

...