The point of modeling to reads is to get the fragment length. You can run MACS without modeling (see nomodel option here) and input a set fragment length (note that fragment length is different from read length). How MACS works is by shifting reads by (fragment length)/2 for forward and reverse (see MACS paper if you are unsure about what I am talking about). I am not sure how applicable this is for Ion Torrent but you can see if you have the same forward/reverse peaks using tag-autocorrelation methods and estimate fragment length that way. A program that can generate data for tag-autocorrelation and estimate fragment length is homer

some relevant links to discussion below:

• http://groups.google.com/group/macs-announcement/msg/d6595465a1f9b212
• http://www.biostars.org/post/show/18548/why-does-chip-seq-data-have-lots-of-reverse-complement-pairs-of-reads/