Brandon:LabNotes/Project1/2012-3-6

From ZhangLabWiki
Revision as of 02:11, 7 March 2012 by >Bsos (Created page with '==peak calling software== *peakseq seems okay *try out macs and compare. macs seemed to call the most peaks, in papers found [http://genome-tech.ucsd.edu/LabNotes/index.php/Bran…')
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to navigation Jump to search

peak calling software[edit]

  • peakseq seems okay
  • try out macs and compare. macs seemed to call the most peaks, in papers found peak calling algorithms comparisons
    • all peaks found by macs were found by the other algorithms. peaks in other programs were not found by macs. false positives?
  • Though peakseq seems less stringent, didn't call many significant peaks when control and sample are analyzed. Also probably many more false positives.

macs14 versus peakseq[edit]

  • macs14 features
    • single sample and sample/control analysis possible
    • macs14 also gives negative peaks (ones with more reads in controls)
    • macs14 outputs files in easier to retrive/analyze format
    • outputs usable bed files (use with bedcoverage or to visualize on UCSC)
    • macs functions better with smaller datasets/reads amounts (then peakseq)
    • gives fold difference in number of reads in the comparison
  • analysis with nextera samples 1 as sample and 4 as control were done, 500 cells sample
    • all peaks called in peakseq were found on macs14.
    • peakseq found 31 peaks, macs14 found 637 peaks.
    • validated peaks in UCSC genome browser with BED file. All with FDR of 0 had more reads in the undigested sample than protease digested control. did with peaks that had least fold different, and most significant peaks.
  • used bedcoverage
' peakseq macs
peaks_called 31 901
bases_covered 18007 539451
macs_onto_peakseq_500_cells 100%
peakseq_onto_macs_500_cells 3.33%

analysis only with macs14[edit]

  • did macs14 analysis on raw UCSC GM12878 data and nextera tagmentation samples 1 and 4.
macs14 peak calls (macs only anslysis) ' ' ' ' '
covered_bp total_bp % covered
ucsc_onto_500cells 182515 539451 33.83347144
500cells_onto_ucsc 182515 82252260 0.221896638
.
peakseq peak calls (peakseq only analysis)
. covered_bp total_bp % covered
Peakseq done on raw UCSC data UCSCv1 UCSC v 500 cells 2115 18007 11.74543233
Peakseq done on raw UCSC data UCSCv2 UCSC v 200 cells 1022 6718 15.21286097
Peakseq done on raw UCSC data UCSCv3 UCSC v 100 cells 1263 7047 17.92252022
  • macs seems to call peaks of the same peaks, have 33% of the same bases covered when ucsc peak calls are mapped onto 500 cell peak calls.(probably consistently open peaks in lymphoblasts)
  • even if dataset is reduced for mapping 500 cell peak calls into UCSC peak calls, % covered would still be .2% to .3%. shows that datasets are very different.
    • this differs from peakseq calls, which had 11% same peak areas called. probably more non specific peak calling. non specific peak calling has been shown in other papers.
  • compare to previous peakseq data analysis


conclusions[edit]

  • as seen before seems that with nextera data, the cell lines are different and thus wont have much reproducibility in terms of accessibility.
  • macs calls more peaks that are accurate, since 33% of those called in 500 cell accessibility data are present in UCSC data.
  • all calls in macs were in peakseq. but only a small number of calls in peakseq were in macs.

future[edit]

  • will read macs14 paper to get better understanding.
  • peakseq can be used as vaidation for top peaks.
  • Useq is comparable to macs14, could also test that out but macs seems best right now. still should look at other programs
  • can do replicates analysis as done here: previous peakseq data analysis