Ns126:Calendar/NOTES/2016-2-1
Jump to navigation
Jump to search
Cancer Specific Methylation haplotype (Novel idea)
- From Dr. Zhang
- We are sequencing the plasma samples with very limited input DNA, equivalent to only a few hundred copies per sample. With the additional major loss of material during bisulfite conversion and library construction, we are probably sampling one or two dozens of molecules per sample. On the other hand, the fraction of tumor DNA in plasma is typically in the range of 0.1% or lower. So if there are 1,000 regions that has tumor-specific haplotypes at 0.1% level and we are looking at 10 unique molecules per region, we would expect to see one tumor haplotypes in only 10 of 1,000 regions. If you move to the next sample, you would see another 10 regions. The chance of finding a set of common regions where we can consistently detect tumor haplotypes is extremely low, at least at the sample size of <100 that we have.
- Along this line of thoughts, random forest is not going to be the right approach for such data. Our best bet is to compare data from tumor tissues and normal plasma, identify a set of genomic regions that are clean in plasma and contain certain level of methylated haplotypes in tumor, then ask whether these regions have an elevated level of MHL.