Dear All,
I would like to ask a question regarding the usage of MACS default parameters for histone modifications for ChIP-Seq data am interrogating. I am running both MACS1.4 and MACS2 on a large cohort of samples , basically for different histone marks like K27me3,K4me1,K4me3 and K27ac. In my case I have treatment and control but still I tried to use the MACS model to see what is the predicted fragment length and the number of peaks obtained by MACS(both by 1.4 and 2). Usually I have seen that people use --nomodel parameter to bypass MACS modelling but from the Feng et al, 2012 paper (Nature Protocols ,2012) what I understood is if the MACS predicted fragment length is less then it is better to go for --nomodel parameter as written in the paper. Since it correlates to much less coverage. In my samples most of the time MACS predicted fragment length have been between 200-350 bp and creating high number of peaks. Strangely for 4 samples the predicted fragment length have been between 51-53 bp however it gave high number of peaks but for one the number of peaks is quite low. Below is the metrics. I am thinking on now running MACS on all these 4 samples (both 1.4 and 2) using --nomodel parameter, even if for some I have high number of peaks. Since the predicted fragment length is too low. What is the understanding of using --nomodel. I could not find anything more than this of using --nomodel parameter. If someone has some better understanding kindly elucidate or if am I thinking it the wrong way. I would like to have some suggestions from experts who have more experience since I have recently started with ChIP-Seq analysis.
Samples MACS1.4_predicted_FL MACS2_predicted_FL MACS1.4_peaks MACS2_peaks
Ascites_K4me1_macs14_out_peaks.bed 52 52 158975 61835
Ascites2_K4me1_macs14_out_peaks.bed 51 51 86460 21317
Tumor1_K27me3_macs14_out_peaks.bed 53 53 72299 45054
Tumor1_K4me1_macs14_out_peaks.bed 52 52 457 38
Do you have any idea of the fragment size from the Bioanalyzer traces? Do you have input data for each sample?
My suspicion is that sample "Tumor1_K4me1_macs14_out_peaks.bed" ChIP has not worked very well and the profile is flat like input. You could check whether it is flat or peaky by comparing your data to ENCODE ChIP-seq and input for the same mods with UCSC browser (try chr22 in wig format) or in IGV.