Kun:LabNotes/MONOD/2014-10-1

From ZhangLabWiki
Jump to navigation Jump to search

Analysis of BSPP_WGBS data[edit]

  • For targeted bisulfite sequencing of cfDNA, we decided to first make WGBS libraries to enrich the short cfDNA fragments, then perform BSPP capture on the amplified libraries.
  • Noi did one round of experiment on 16 samples, and the libraries were sequenced on multiple HiSeq Rapid Runs.
  • Dinh ran the mapping of the 140914_HiSeqRapidRun data, and the results are in /media/Ext12T/DD_Ext12T/MONOD/140914_HiSeqRapidRun/NO_RMDUP.
  • To compare with my previous BSPP results, I also did the mapping of 140919_HiSeqRapidRun data in the old way, and keep the results in /home/kunzhang/CpgMIP/MONOD/Data/140919_HiSeqRapidRun/SE150. Note that the same batch of libraries were sequenced in multiple flow cells, so the results can be combined or compared.
Sample id Input All mapped reads On-target reads On-target rate
6-T-2 gDNA 15,259,826 12,294,400 80.6%
6-T-4 gDNA 16,834,556 15,695,043 93.2%
6-T-3 gDNA 15,454,340 14,637,901 94.7%
COLO2-03 WGBS 13,098,434 571,663 4.4%
NORM1-03 WGBS 13,383,336 649,907 4.9%
cfDNA WGBS 9,498,520 585,118 6.2%
NORM1-02 WGBS 12,839,754 853,671 6.6%
6-P-5 WGBS 14,116,370 1,201,083 8.5%
6-P-4 WGBS 10,335,816 982,269 9.5%
6-P-2 WGBS 9,200,684 1,352,128 14.7%
6-P-3 WGBS 12,153,832 4,051,723 33.3%
NC-30 WGBS 9,341,122 4,873,918 52.2%
NC-27 WGBS 8,709,818 5,455,158 62.6%
NC-9 WGBS 10,843,616 7,406,900 68.3%
NC-6 WGBS 18,171,498 15,146,583 83.4%
  • Clearly there is a hug variability in terms of mapping rates. The libraries from tumor gDNA have the highest mapping rate. Some libraries from plasma WGBS libraries also have decent mapping rates, but six libraries have very low mapping rates.
  • Next I looked into absolute probe capturing efficiencies. Since there is the clear difference between gDNA or WGBS libraries as the input, I split the samples into several groups.
        Efficiency       # probes
          >0.05               27
          >0.01            1,158
          >0.005           2,884
          >0.001          10,704
          >0.0005         15,209
          >0.0001         24,771
          >0              35,272
          Total           39,722
    • Based on the data from gDNA, the UMI counts per probe are quite consistent across the four samples. Among the most efficient 2884 probes, the average CV is 0.18. Also for 100ng of input DNA, roughly 18,440 probes are usable for quantifying methylation (unique counts >=10).
    • In contrast, the capture efficiencies are highly variable among the WGBS libraries. The average CV for the top 2884 probes is 1.60, or roughly 10x more variable than gDNA. What explain such variability?
      • Noi hypothesized that the insert sizes for ctDNA WGBS libraries center at around 180bp. The target size of our BSPP probes is in the range of 100-150bp, and the annealing arms are roughly 50bp in length. So some probes of large gap size might not capture well on short molecules in the libraries. I did observe a significant positive correlation (R=0.26) between the CV of capturing efficiencies on plasma and the insert size. So this is one of the reason, but still it only explain a very small fraction.
      • Interestingly, the average CVs for the six "good" plasma libraries and the six "poor" plasma libraries are different and both smaller than the CV for 12 plasma samples combined, with the six "poor" plasma libraries being the lower (average CV=0.96). So this is an indication that the variability has something to do with the libraries.
  File:Extreme variability of BSPP on WGBS 02Oct2014.png