Any way to remove background reads?

[metheuse]

Hi, I'm dealing with some ChIP-seq data which is quite noisy. MACS can only find less than 1k peaks. I checked its cross-correlation. The NSC is 1.02 and the RSC is 0.22. Based on ENCODE recommendation, these are "bad".
I'm just wondering if it's possible to remove those reads which form higher "read length" correlation than "fragment length" correlation, to reduce the phantom peak?
Btw, I also checked the nonduplicated read fraction (=0.82 which is good), and the fraction of mapped reads in peaks (=7% which passes ENCODE's 1% metrics, though doesn't necessarily mean a good one.)
I looked at the bigwig signals on browser. The signal does look quite noisy. Even in the peaks MACS called, it doesn't appear to be real peaks.
Is there any way to rescue this kind of data bioinformatically?
Thanks for any suggestions.

[asurarocks]

Hi metheuse,
Did you figure it out? I am dealing with a noisy medip-seq data and I wish to remove the reads that are contributing to noisy peaks. Any help will be much appreciated!
Thanks,

[metheuse]

Hi metheuse,
Did you figure it out? I am dealing with a noisy medip-seq data and I wish to remove the reads that are contributing to noisy peaks. Any help will be much appreciated!
Thanks,

Nope, I didn't get any useful information.

[asurarocks]

Here is what I came up with:
Convert BAM/SAM to BED (narrowpeak), find peaks using SPP/MACS, set a cutoff for noisy peaks and obtain their intervals, use BEDTools to subtract those intervals from the original BED file (you can convert them back to BAM/SAM if needed, I guess). I know there should be some better way to do this, but this is as far as I can go with my limited programming knowledge. I hope this helps.