(Galaxy) MACS2 peaks threshold

Question

[Morris_Chair]

Hello everyone,

I am using galaxy to analyze ChIP-seq data downloaded from SRA. This team used as negative control the IgG and when I compare my genes of interest against the control using IGV I can visually see an enrichment of reads nearby the genes but MACS2 does't not recognize this peaks as statistical significant.

is it possible to lower the MACS2 threshold and perhaps this peaks become significant? I think one of the disadvantage of using IgG is that few proteins are pulled down anyways

Thank you

[GenoMax]

If the question is about how to do this in galaxy then you may want to post this at Galaxy help site.

[ATpoint]

I think one of the disadvantage of using IgG is that few proteins are pulled down anyways

This is exactly why you do IgG, which is to determine if your protein might be pulled down simply by unspecific antibody binding. I would leave things at default, and strongly recommend against eyeballing peaks. If it is not significant, there is nothing you can do about it. THe upper panel looks quite noisy so I (based on my experience with good IPs) would not call any of those "peaks" significant by eye.

Yes, you can of course lower the threshold, but again, I would not do it. You only end up with a lot of false-positives.

[Morris_Chair]

Hi ATpoint, in part I agree with you but I think that if it was input instead of IgG it would have been different, less background because reads are more randomly distributed (I guess). I adjusted the parameters as recommended on a galaxy tutorial that I found on youtube, do you think is it better to leave as default like that?

regardless of fragment size and other parameters?

I think this parameters are still too stringent, not even the regions mentioned in the paper from where this data come from are significant with MACS2. Probably I see few peaks because there is not a good depth of sequencing?

Thanks

[ATpoint]

Look, if it is not significant, it is not significant. There are a lot of sub-optimal quality data out there and I doubt you gain anything from adjusting parameters until you get the desired result. The IgG is the better control as it controls for unspecific binding. Did that paper do anything to verify that region beyond NGS? Being not able to reproduce the finding is rather an argument against that dataset. I've seen datasets which basically had no callable peaks except a few (like less than 10) and the paper then still praised these 10 peaks in fancy screenshots in the main figure, suggesting a high-quality ChIP which it simply wasn't. Bottom line: I personally would not tweak parameters to get things significant. The defaults are good and real peaks should come out with this. Everything else would need external experimental validation.

[Morris_Chair]

yes they validated with qPCR later, Thank you for your answer it helps :)

[ATpoint]

By the way, what you are looking at in the browser are bam files right? You have to normalize them, in the very least for sequencing depth, to make conclusions based on visual inspection. Check bamCoverage from deeptools and its normalization options.

[Morris_Chair]

yes in the picture is a bam, bamCoverage doesn't like my code and I have error,

bamCoverage -bam SRR1104880.bam -binSize 10 -ignoreDuplicates -normalizeTo1x 2864785220 -fragmentLength 200 -o Coverage.SRR1104880.bw

there is a conflict with Numpy :/

I see that galaxy has a similar function to normalize and compare two bigwig files too, else I can use other codes to remove the reads that exceed isn't it?

macs2 randsample -t  myfile.bed -n 2864785220 -o fixedfile.bed

thanks