Dinh/Dinh 2014/NOTES/2014-4-1: Difference between revisions

From ZhangLabWiki
Jump to navigation Jump to search
>Dinh
mNo edit summary
>Dinh
Line 54: Line 54:
* shuffle sample labels
* shuffle sample labels
=== Next steps ===
=== Next steps ===
* Determine whether method 1, 2 or 3 is more appropriate. They seems to have very similar outcomes, which is encouraging.
* Determine whether method 1, 2 or 3 is more appropriate.
** Generate plots to see whether the DMRs are also detectable by eye.
** Generate plots to see whether the DMRs are also detectable by eye.
* Does the lower depth of coverage for the non-ESC samples (Habibi et al data have 30x average depth of coverage while our data have between 7-10 average depth of coverage) affect the number of DMRs detected for iPSCs versus SCNTs ? I need to downsample the Habibi and Stadler data.
* Does the lower depth of coverage for the non-ESC samples (Habibi et al data have 30x average depth of coverage while our data have between 7-10 average depth of coverage) affect the number of DMRs detected for iPSCs versus SCNTs ? I need to downsample the Habibi and Stadler data.
* Use MOABS to find DMRs between 2i ESC and 2i SCNTs
* Use MOABS to find DMRs between 2i ESC and 2i SCNTs

Revision as of 01:39, 5 April 2014

DMR finding with MOdel-based Analysis of Bisulfite Sequencing data (MOABS)

  • Using only the mcomp module from the MOABS pipeline requires:
 1. setting LC_ALL variable with:
    export LC_ALL=C
 2. installing RInside for R. Open R and type:
    install.packages("RInside")
  • The input for mcomp is a G.bed file (ie), more detailed explanation see the MOABS documentations.
 #chrom	start	end	ratio	totalC	methC	strand	next	Plus	tcP	mcP	Minus	tcM	mcM
 chr5	3009585	3009587	1	1	1	-	G	+	0	0	-	1	1
 chr5	3011471	3011473	0.711	45	32	B	G	+	28	18	-	17	14

Analysis of mouse WGBS

  • Prepare the G.bed files with methylFreq2Gbed.pl (genome-miner)
 /home/dinh/scripts/methylFreq2GBED.pl 1 < [methylFreq] > [BED name]
  • We perform 3 tests:
 esc versus ipsc
 esc versus scnt
 ipsc versus scnt
  • MOABS commands:
 export LC_ALL=C
 /home/dinh/softwares/moabs-v1.2.7.src.x86_64_Linux.data/bin/mcomp -r merged.esc.$c.G.bed -r merged.scnt.$c.G.bed -c moabs.esc_scnt.$c -l esc.$c scnt.$c
 /home/dinh/softwares/moabs-v1.2.7.src.x86_64_Linux.data/bin/mcomp -r merged.esc.$c.G.bed -r merged.ips.$c.G.bed -c moabs.esc_ips.$c -l esc.$c ips.$c
 /home/dinh/softwares/moabs-v1.2.7.src.x86_64_Linux.data/bin/mcomp -r merged.ips.$c.G.bed -r merged.scnt.$c.G.bed -c moabs.ips_scnt.$c -l ips.$c scnt.$c

Results

  • MOABS calculated DMRs using 3 methods. Need to look up what the three methods are..
  • Method #1, # DMRs
    • call Dmc and Dmr according to method 1: Fisher Exact Test
    • Dmc is defined as depth >= minDepth(suggest 10), p<p_fet_cut(suggest 0.05), nMethDif > nominalMethDifCutoff(0.3333)
    • Dmr is defined as minC >=3, maxDist 300, p<p_fet_cut(suggest 0.05), nMethDif > nominalMethDifCutoff(0.3333)
 22,125 esc_vs_ips File:Dmr M1 esc vs ips.txt
 22,135 esc_vs_scnt File:Dmr M1 esc vs scnt.txt
    709 ips_vs_scnt File:Dmr M1 ips vs scnt.txt
 
  • Method #2, # DMRs
    • call Dmc and Dmr according to method 2: Using the credible methylation difference metric (beta-binomial) distribution.
    • Dmc is defined as cMethDif > credibleMethDifCutoff(0.2)
    • Dmr is defined as minC >=3, maxDist 300, cMethDif > credibleMethDifCutoff(0.2)
 19,799 esc_vs_ips File:Dmr M2 esc vs ips.txt
 15,318 esc_vs_scnt File:Dmr M2 esc vs scnt.txt
     53 ips_vc_scnt File:Dmr M2 ips vs scnt.txt
  • Method #3, # DMRs
    • call Dmc according to method 2
    • Dmc is defined as cMethDif > credibleMethDifCutoff(0.2)
    • call Dmr using Hidden Markov Model
 20,124 esc_vs_ips File:Dmr M3 esc vs ips.txt
 15,550 esc_vs_scnt File:Dmr M3 esc vs scnt.txt
     21 ips_vc_scnt File:Dmr M3 ips vs scnt.txt
  • The SCNTs appear to be more similar epigenetically to iPSCs than to ESCs.
  • I converted the DMR files to bed format and uploaded them to GREAT for enrichment analysis. For both sets of DMRs between esc and ips/scnt, the pathway that comes up is protein citrullination.
  • I looked up the term and found an interesting excerpt from Wikipedia:
    • Citrullination is important because it controls the expression of genes, particularly in the developing embryo, and because the immune system often attacks citrullinated proteins, leading to autoimmune diseases such as rheumatoid arthritis and multiple sclerosis.

False discovery rate

  • which method have the lowest false discovery rate?
  • shuffle sample labels

Next steps

  • Determine whether method 1, 2 or 3 is more appropriate.
    • Generate plots to see whether the DMRs are also detectable by eye.
  • Does the lower depth of coverage for the non-ESC samples (Habibi et al data have 30x average depth of coverage while our data have between 7-10 average depth of coverage) affect the number of DMRs detected for iPSCs versus SCNTs ? I need to downsample the Habibi and Stadler data.
  • Use MOABS to find DMRs between 2i ESC and 2i SCNTs