Dinh/Dinh 2015/NOTES/2015-10-12

From ZhangLabWiki
Jump to navigation Jump to search

Genome segmentation - tissue specific CpG sites[edit]

Question: how to identify regions with high methylation variabilities in unbiased way?[edit]

  • No hard cutoffs - utilize HMM segmentation to identify regions
  • No coverage bias - regions with low coverages have more noise - utilize BSmooth to perform smoothing over genomic regions to "boost" up the confidence of methylation measurements

Question: which metric of methylation variability should be used?[edit]

  • Coefficient of variation - standard deviation / mean = high values for low mean. This may not be altogether bad, since low mean could means that there are mostly unmethylated tissues and few methylated tissues. High CV regions are highly enriched for DHS and TFBS.
  • Standard deviation - too variable?
  • Maximum absolute pairwise distance - definitely show which region have tissue variable methylation when this value is high
  • Solution: utilize a combination of high coefficient of variation and high maximum absolute pairwise distance to identify a set of high methylation variable sites.
  • After filtering out sites categorized as high coefficient of variation and high maximum absolute pairwise distance - 2,562,875 CpG sites (10.6% out of 24,224,775 included in the segmentation)
  • Combined sites within 100 bp of each other into one region - 646,294 regions
  • 646,294 regions
 Minimum size: 1 bp, Median size: 38 bp, Mean size: 92.85 bp, Max size: 4,544 bp
  • I filtered out sites within the 95% percentile of the low or middle segmented states:
 95%-tile of max absolute distance : 0.339
 95%-tile of coefficient of variation: 0.142412
 Remaining tissue specific CpGs : 2,557,584
  • Combined sites within 100 bp of each other into one region - 644,917

Question: how to reduce down to a set of 5hmC-relevant sites?[edit]

  • We want to validate a small subset of 5hmC relevant sites - utilize Tet protein binding data
  • Tet protein occupancy have been found for two cell lines (colon cancer and embryonic kidney)
 TET2 HCT116 - 2,536 significant peaks - 612 peaks overlap with HEK293T - 785 (31.0% overlap with high variance regions)
 TET2 HEK293T - 4,270 significant peaks - 612 peaks overlap with HCT116 - 3112 (72.9% overlap with high variance regions)
 TET3 HEK293T - 4,983 significant peaks - 2952 (59.2% overlap with high variance regions)
  • Since these TET protein occupancy data are cell type specific, we may be limited in what we can study. Perhaps just in colon cancer or kidney
  • Overlap all three Tet2 occupancy files, found 13,530 regions to target.

Question: where are the 644,917 regions?[edit]

  • I extended each region by 50 bp upstream and downstream, so the minimum region size will be 100 bp long.