Dinh/Dinh 2012/NOTES/2012-2-13

From ZhangLabWiki
Jump to navigation Jump to search

24 ChIP-seq - Stimulated and Resting Neurons[edit]

  • Analysis at repeats.
Downloaded RepeatMasker track fro mm9. Then filtered and get Non-overlapping windows:
 2072  awk '{if($12 ~ /DNA/) print $0}' RepeatMasker_mm9.txt | ./getNonOverlapping.pl > DNA_repeats.nonoverlapping &
2073  awk '{if($12 ~ /LINE/) print $0}' RepeatMasker_mm9.txt | ./getNonOverlapping.pl > LINE_repeats.nonoverlapping &
2074  awk '{if($12 ~ /SINE/) print $0}' RepeatMasker_mm9.txt | ./getNonOverlapping.pl > SINE_repeats.nonoverlapping &
2075  awk '{if($12 ~ /LTR/) print $0}' RepeatMasker_mm9.txt | ./getNonOverlapping.pl > LTR_repeats.nonoverlapping &
2076  awk '{if($12 ~ /Satellite/) print $0}' RepeatMasker_mm9.txt | ./getNonOverlapping.pl > Satellite_repeats.nonoverlapping &
File:GetNonOverlapping.txt
  • Note that some repeat segments overlaps between these difference repeat class.
cat *_repeats.nonoverlapping | sort -k1,1 -k2,3n > sorted.Repeats (3,188,316 entries)
./removeOverlapping < sorted.Repeats > tmp
mv tmp sorted.Repeats (3,187,848 entries)
File:RemoveOverlapping.txt
  • Remove chr*_random entries
awk '{if($1 !~ /_random/) print $0}' sorted.Repeats > sorted.Repeats_norandom
  • Final repeat regions file: 3,127,206 entries
Started with 3,188,316 non-overlapping entries for DNA, LINE, SINE, LTR, and Satellite repeats
Then removed entries/segments which have over-lapping labels -> 3,187,848 entries left
Then removed entries on chr*_random -> 3,127,206 entries -> Non-overlapping repeat regions
  • Sort BED and remove duplicates
for f in Indx*.BED 
do
       sort -k1.4,1.4n -k2,2n -k3,3n -u -T tmp $f > rmdup.$f
done;
  • Strand specific normalization with both IgG and Input controls. Normalization of read depth is taken care of by regression model.
ftumch2 -i rmdup.Indx1.BED -c rmdup.Indx7.BED rmdup.Indx4.BED -o normalized.R.cac.BED -rl 350 -cs 100000 -ws 5 -iv 5
#ftumch2 -i rmdup.Indx2.BED -c rmdup.Indx7.BED rmdup.Indx4.BED -o normalized.R.fc.BED -rl 350 -cs 100000 -ws 5 -iv 5
#ftumch2 -i rmdup.Indx3.BED -c rmdup.Indx7.BED rmdup.Indx4.BED -o normalized.R.hmc.BED -rl 350 -cs 100000 -ws 5 -iv 5
#ftumch2 -i rmdup.Indx5.BED -c rmdup.Indx7.BED rmdup.Indx6.BED -o normalized.R.mc.BED -rl 350 -cs 100000 -ws 5 -iv 5
#ftumch2 -i rmdup.Indx8.BED -c rmdup.Indx14.BED rmdup.Indx11.BED -o normalized.S.cac.BED -rl 350 -cs 100000 -ws 5 -iv 5
#ftumch2 -i rmdup.Indx9.BED -c rmdup.Indx14.BED rmdup.Indx11.BED -o normalized.S.fc.BED -rl 350 -cs 100000 -ws 5 -iv 5
#ftumch2 -i rmdup.Indx10.BED -c rmdup.Indx14.BED rmdup.Indx11.BED -o normalized.S.hmc.BED -rl 350 -cs 100000 -ws 5 -iv 5
#ftumch2 -i rmdup.Indx12.BED -c rmdup.Indx14.BED rmdup.Indx13.BED -o normalized.S.mc.BED -rl 350 -cs 100000 -ws 5 -iv 5
#ftumch2 -i rmdup.Indx15.BED -c rmdup.Indx19.BED rmdup.Indx18.BED -o normalized.R.dnmt3a.BED -rl 350 -cs 100000 -ws 5 -iv 5
#ftumch2 -i rmdup.Indx16.BED -c rmdup.Indx19.BED rmdup.Indx18.BED -o normalized.R.tdg.BED -rl 350 -cs 100000 -ws 5 -iv 5
#ftumch2 -i rmdup.Indx17.BED -c rmdup.Indx19.BED rmdup.Indx18.BED -o normalized.R.tet2.BED -rl 350 -cs 100000 -ws 5 -iv 5
#ftumch2 -i rmdup.Indx20.BED -c rmdup.Indx24.BED rmdup.Indx23.BED -o normalized.S.dnmt3a.BED -rl 350 -cs 100000 -ws 5 -iv 5
#ftumch2 -i rmdup.Indx21.BED -c rmdup.Indx24.BED rmdup.Indx23.BED -o normalized.S.tdg.BED -rl 350 -cs 100000 -ws 5 -iv 5
#ftumch2 -i rmdup.Indx22.BED -c rmdup.Indx24.BED rmdup.Indx23.BED -o normalized.S.tet2.BED -rl 350 -cs 100000 -ws 5 -iv 5


BEDtools[edit]

  • Used bedTools to convert all my BED (mapped reads) to bedGraph format (this is for validation on raw data)
for f in rmdup*BED
do
       genomeCoverageBed -bg -i $f -g mm9.fai > $f.bdg
done;

SISSRs[edit]

  • Peak calling using SISSRs (a simple perl script)..FDR calculated from a Poisson distribution.
for f in normalized*.BED
do
       sissrs.pl -i $f -s 2654911517 -m 0.7 -F 350 -u -D 0.001 -o $f.peaks
done;
-i (input BED file)
-s (genome size in bp)
-m (fraction of genome covered, estimated 0.7)
-F (average fragment length, inferred from sampe.log (BWA) the median is 300, and average could be 350.
-u (report peaks which is not supported by at least two reads in both directions as well)
-D (FDR, 0.001)
-o (output peak file)
  • Peaks for properly paired reads:
  8131 normalized.R.cac.BED.peaks
   474 normalized.R.dnmt3a.BED.peaks
  7418 normalized.R.fc.BED.peaks
  8838 normalized.R.hmc.BED.peaks
 21352 normalized.R.mc.BED.peaks
   565 normalized.R.tdg.BED.peaks
   806 normalized.R.tet2.BED.peaks
  7300 normalized.S.cac.BED.peaks
   653 normalized.S.dnmt3a.BED.peaks
  5653 normalized.S.fc.BED.peaks
 33480 normalized.S.hmc.BED.peaks
 32660 normalized.S.mc.BED.peaks
   548 normalized.S.tdg.BED.peaks
   704 normalized.S.tet2.BED.peaks
  • Peaks for improperly paired reads
 (to be done)

Validation[edit]

  • I uploaded the bedGraph files to genome browser to validate the peaks.
  • The chIP peaks (despite high number of reads for all chIP samples) had a lower peak validation rate (~50%). About half the times, I would see many peaks across all the controls at the regions where a peak was called, but the peak for the treat data was only slightly higher.
  • The dIP peaks had good peak validation rate.

FindPeak 4.0[edit]

  • Peak calling using FindPeak 4.0. The reason why MACS1.4 wouldn't work because MACS14 "shifts" read toward the center when reads are pileup on the forward strand and reverse strand. This shifting model wouldn't work for finding enrichment of certain regions in the genome (unlike motif finding, the size of the region is unknown and may be very large).
for f in normalized.*BED
do
       awk '{print $1"\t"$2"\t"$3"\t"$4"\t0\t"$6}' $f > tmp
       java -Xmx2G -jar /usr/local/fp4/FindPeaks.jar -input tmp -aligner bed -output . -wig_step_size 1 - dist_type 0 350 -subpeaks 0.5 -landerwaterman 0.001 -name $f -one_per
done;
  • Number of peaks called for each dataset:
  41109 normalized.R.cac.BED_fixed_350_subpeaks.peaks
    820 normalized.R.dnmt3a.BED_fixed_350_subpeaks.peaks
  12382 normalized.R.fc.BED_fixed_350_subpeaks.peaks
  14825 normalized.R.hmc.BED_fixed_350_subpeaks.peaks
  28826 normalized.R.mc.BED_fixed_350_subpeaks.peaks
   1301 normalized.R.tdg.BED_fixed_350_subpeaks.peaks
   2177 normalized.R.tet2.BED_fixed_350_subpeaks.peaks
  61289 normalized.S.cac.BED_fixed_350_subpeaks.peaks
   1431 normalized.S.dnmt3a.BED_fixed_350_subpeaks.peaks
   9107 normalized.S.fc.BED_fixed_350_subpeaks.peaks
  51385 normalized.S.hmc.BED_fixed_350_subpeaks.peaks
  52640 normalized.S.mc.BED_fixed_350_subpeaks.peaks
   1222 normalized.S.tdg.BED_fixed_350_subpeaks.peaks
   1090 normalized.S.tet2.BED_fixed_350_subpeaks.peaks
  • Things to note: For all properly paired reads there are fewer peaks for ChIP data than for dIP data. This is expected since the ChIP are peaks for DNA binding proteins (they are specific). Whereas for the dIP data, it is possible to have more signals since there are many location for cytosines to be oxydized.
  • I also used the same normalization and peak calling method (except with dist_type 0 110) for the improperly paired reads.
 16759 normalized.R.cac.BED_fixed_110_subpeaks.peaks
    95 normalized.R.dnmt3a.BED_fixed_110_subpeaks.peaks
  4250 normalized.R.fc.BED_fixed_110_subpeaks.peaks
  1720 normalized.R.hmc.BED_fixed_110_subpeaks.peaks
  2047 normalized.R.mc.BED_fixed_110_subpeaks.peaks
    94 normalized.R.tdg.BED_fixed_110_subpeaks.peaks
   130 normalized.R.tet2.BED_fixed_110_subpeaks.peaks
 16148 normalized.S.cac.BED_fixed_110_subpeaks.peaks
   112 normalized.S.dnmt3a.BED_fixed_110_subpeaks.peaks
  1688 normalized.S.fc.BED_fixed_110_subpeaks.peaks
  1097 normalized.S.hmc.BED_fixed_110_subpeaks.peaks
  1212 normalized.S.mc.BED_fixed_110_subpeaks.peaks
   109 normalized.S.tdg.BED_fixed_110_subpeaks.peaks
   104 normalized.S.tet2.BED_fixed_110_subpeaks.peaks
  • Things to note: much fewer peaks found for ChIP data, which is expected, since DNMT3a, TDG, and Tet2 are not known to bind in repeats. Up to 10x greater signals are observed for caC marks in both resting and stimulated neurons.