Generate appropriate background

  • I previously used within-sample DMRs filtered for coverage in 3/4 individuals/group as the background.
    • HOWEVER, these did not include all sites that had the potential to be methylated
    • To create a more inclusive background, I need to look at all CG sites considered prior to determining within-sample DMRs
  • Within-sample DMRs were determined by:
    • having at least 3 differentially methylated sites (DMS)
      • DMS were determined by:
        • having 5x coverage
          • if sites are within a 30bp window their counts can be combined
        • passing an RMS test significance threshold of 0.01
    • within a max distance of 250bp
  • ATTEMPT 1: If I remove the significance threshold (and set –sig-cutoff to 1 instead of 0.01) for DMS, than I should get results from all sites considered for DMS.
    • I attempted this using this mox script: 20191108_DMRfindAllEPInoSig.sh
      • RESULTS: Resulting files indicate some other filtering is happening because because the number of regions is the nearly the same as when I ran DMRfind with significance level set to 0.01. I had expected a much longer list of regions.
  • ATTEMPT 2: To address if the software only consider CG sites with a value in the methylation counts column of the allc file I created new allc files with the coverage counts column also as the methylation counts column (assuming 100% methylation at every site).
  • ATTEMPT 3: To adjust other parameters in attempt to remove all filtering, I tried the following settings in this mox script here: 20191108_DMRfindAllEPItotCountsRes1.sh on ambient samples only as a test:
    • –resid-cutoff 1
    • –min-tests 1
    • –num-sims 1
    • –sig-cutoff 1
    • RESULTS: resulting files indicate certain criteria in the software are not met as output contains only a header

4th times the charm!

Plot background vs. sig. DMR features

I updated the Rmarkdown file I used previously to generate a bar plot showing the proportion of features that significant DMRs vs. background sites fall into.

RESULTS: There are no strong differences between significant DMRs and background CpG sites in the proportion of features they overlap with, except:

  • Day145 comparison where CDS and exon features are under-represented, and repeat_regions are over-represented.

  • CDS and exon are slightly over-represented in Day135 and all ambient sample comparisons

Next steps

  • Look deeper into repeat regions (there are ~6 different catagories, so can see if any are particularly different)
  • For DMRs no in features, check the nearest features
  • Compare genes with DMRs to the ones identified by Hollie’s method
  • Continue GO analysis
    • generate appropriate GO background to use for each comparison
    • run TopGO
  • Go through manuscript methods section
    • update new stuff
    • add comments to areas of concern