When I perform cuttag, I used IgG primary antibody as the control group, but its overall background signal was very high. I guess the quality of the antibody was not good. Is there any other reason for this?
So those deeptools plots don't really contain the information of how "high" your background is. They give you the relative enrichment (coverage) of genic positions, scaled to the overall number of reads (or fragments). If your target profile looks like your IgG profile, you can be reasonably sure that you have "high background" or "low signal." Maybe I see one shared blip at ~2.1kb down from the TSS, but this kind of suggest that the signal to noise ratio is like 4:1 or 5:1. Not excellent, but also pretty good.
More pertinent questions are:
(1) What were the concentrations (ng/ul) of the libraries? Target should be N ng/ul and IgG something like N/10 or N/20 ng/ul.
(2) What is the FRiP/FFiP of the target? This should be >50% for a good Cut&Tag run; but ChIP can be ~30% and that's workable.
(3) What do the duplication (% duplicates at 1M, 2M, ..., 50M reads) curves look like? You should see much higher duplications for the IgG library.
Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
version 2.3.6
Great thanks to LChart!
I think the specific question regarding the top post is:
We can't see the clear peak enrichment at TSS in IgG sample, but the average signal levels of IgG (background) is higher as compared with the H3K27ac signal (apparent enrichment at TSS, with lower overall background)
From both the tag density map and the heatmap, we initally assume that there should not be any peak enrichment (or the higher background likely come from the random, non-specific binding,
However, when we took a deeper look, we were surprised to see that the overall higher background for IgG is actually from tens to hundreds of seemingly non-random distributed genomic sites.
This is strange to us. Any tips as to avoiding this overall background signal for IgG is greatly appreciated.
TSS are not random. They're open chromatin, hence enriched for protein binding, and that will attract more IgG binding than in orphan regions across the genome with no notable protein enrichment. I would say the IgG track looks perfectly expected. The baseline is indeed a bit higher than in the IP track, but this is almost certainly due to the normalization which is probably just library size scaling. A more sophisticated method would put both baselines at the same level. But since IgG is typically only used for peak calling I wonder whether it matters a lot.
I think the data only shows that you have a successful IP. Hence, call peaks and do downstream analysis. There is not more you can do at this point. I slightly disagree with the good answer from LChart that a good IP must have FRiPs over 50%, that is excessively high in my experience. I saw ChIP-seq (not C&T though) that had FRiPs like 5% and were perfectly usable. Noise is always an issue in these antibody-based assays. Check the data on the IGV, you need to see clear separation between peaks and noise, but based on the IP track here I would absolutely expect that.
Dear ATpoint, thanks for your kind reply.
due to the stringent requirement, could you specifically describe how to perform the "A more sophisticated method would put both baselines at the same level."? because we need to remove all the possible peaks in IgG sample.
Many thanks!
Normalizations like DESeq2, but this is not implemented in peak callers so I don't think this is easily available. In my opinion, IP-based methods or NGS in general is not suited for binary statements (peak or no peak), but more for comparative analysis between conditions. It's like in RNA-seq "is a gene expressed", this is like an unsolved problem ever since, same here "is it a peak or not". Why not just call peaks with standard tools, assess reproducibility between replicates (for example IDR), filter against the ENCODE blacklist, the usual things? You will never get a 100% confident list of peaks, as the transition between signal and noise is continuous and not sharp/discrete. Especially true for H3K27ac because it's often wider peaks, unlike transcription factors that very very distinct narrow signals.