Paired end data: amount of identified peaks depends on peak shift estimate

[timdeme...@gmail.com]

Dear Anton, all,

I have applied QuEST on a set of 10 experiments (4 histone modifications, 1 transcription factor, 2 conditions for each), with a total input control for each condition. Sequencing data is paired-end, insert size approximately 300bp, and I used the histone-type ChIP or transcription factor options as appropriate and the recommended stringent parameter setting.

QuEST yields peak_shift_estimates of ~150nt for 9/10 experiments, which is half of the insert size and therefore most likely a good estimate. However, for these experiments (8 histone mods, 1 TF) I typically find very low amounts of peaks (0 to 2) with very small FDRs. On the other hand, for one experiment (TF), the insert size is estimated at ~60nt, and about 2500 peaks are detected (ca. 500 accepted), but with a very high FDR (>94%).

It appears that the amount of identified peaks heavily depends on the peak shift estimate. Is this correct?
Any advice on how to continue with this experiment? I guess I can overrule the peak shift estimate to the more correct 150bp using the "advanced" settings, but this still leaves me with virtually no "significant" results. As I cannot believe that each ChIP-experiment failed (all QC steps worked out fine), and the modifications are quite common (i.e. far more results are expected), I assume that the parameter settings are not optimized for this kind of paired-end experiment. Any advice?

Thank you!
Best regards,
Tim

[Anton]

Hi Tim,

the number of peaks does not strongly depend on the peak shift for small values of peak shift (<500 bp). The FDR depends on the QuEST peak calling thresholds and the number of detected peaks. It seems like you detected 2,500 peaks when using low threshold, hence high FDR. What settings are you using for peak calling? 

All the best, Anton