Q: How exactly does cutoff-analysis work?

[gdolsten]

Hi, I am using --cutoff-analysis to see where to put my p-values in combination with IDR. I am confused, because I called peaks with default settings:

macs2 callpeak -t ${files[@]} -f BAMPE --call-summits -g mm -B --SPMR --outdir ./peaks/ -n testmacs --cutoff-analysis

When I look at cutoff analysis file, I see this:

9.90	7.85	8170	22838154	2795.37
9.60	7.55	8334	23012621	2761.29
9.30	7.27	8655	24020243	2775.30
9.00	6.98	8991	24975334	2777.81
8.70	6.70	9329	25974511	2784.28
8.40	6.41	9749	27042216	2773.85
8.10	6.13	10163	28149912	2769.84
7.80	5.84	10562	29406424	2784.17
7.50	5.55	10765	29635006	2752.90
7.20	5.27	11255	30944376	2749.39
6.90	4.98	11822	32334821	2735.14
6.60	4.70	12361	33889368	2741.64
6.30	4.42	12917	35486336	2747.26
6.00	4.14	13521	37235498	2753.90
5.70	3.86	14110	39072368	2769.13
5.40	3.58	14803	41062556	2773.93
5.10	3.30	15623	43265761	2769.36
4.80	3.02	16470	45620369	2769.91
4.50	2.74	17372	48154532	2771.96
4.20	2.47	19321	53637710	2776.14
3.90	2.21	20500	57022628	2781.59
3.60	1.93	21886	60767021	2776.52
3.30	1.66	23239	64844893	2790.35
3.00	1.39	26275	73758160	2807.16
2.70	1.14	28250	79378821	2809.87
2.40	0.87	30249	85459624	2825.20
2.10	0.63	34883	99113222	2841.30
1.80	0.38	40607	115671821	2848.57
1.50	0.13	44369	126210817	2844.57
1.20	0.00	52441	149797749	2856.50
0.90	0.00	68738	196269644	2855.33
0.60	0.00	93683	264697132	2825.46
0.30	0.00	134088	369036298	2752.19

However, when I examine the number of peaks with those q-values (or p-values) by filtering the .narrowPeak file, I see many more in my .narrowPeak than the cutoff-analysis claims. Does this mean that cutoff-analysis calls a separate set of peaks for each q-cutoff described above, rather than simply thresholding the peaks?

[taoliu]

However, when I examine the number of peaks with those q-values (or p-values) by filtering the .narrowPeak file, I see many more in my .narrowPeak than the cutoff-analysis claims. Does this mean that cutoff-analysis calls a separate set of peaks for each q-cutoff described above, rather than simply thresholding the peaks?

Yes. cutoff-analysis calls a separate set of peaks for each q-value cutoff. Simply threshold-filtering the peak result won't work in the same way. Because the q-value in the peak result only represents the highest/most significant position inside of the peak.