Paper notes

Paper discussion notes:

• add converted reference to figure 1 -> figure 2
• add rescue component to figure 2, promote to figure 1
• introduction
  • MapQ is routinely used in standard sequencing to avoid variant calls from mixed sites. More important and difficult for methylation assessment - MapQ is not sufficient. Per-C mappability filtering also not sufficient. Need to filter individual reads.

Results

• miscalls are widespread and in interesting regions
• MapQ is not sufficient to filter these out (figure 4)
• MapQ trace Figure 4: should show mappability filtering effect in an interesting gene near in interesting gene

Figure 4a:

• schematic diagram of filtering method

Figure 6: frequency of over and under filtering

Clinical effect figures:

Looking at same samples studied many times - do we see higher variance in

Mixed methylation to figure 1 and introduction .

Simulation: will help us be convinced that purifying methylation observation is real / relevant

• Different loci with same (short? - How long?) sequence but with 100% and 0% methylation (ej locus A=100% and locus B=0%) that show variable % of methylation for all loci without Caiden's mapability filtering (e.g. A=40% and B=60%) but turn correct (we created the correct!) after filtering.

Clinical data:

• tumor samples from methylation biomarker discovery paper
• tumor heterogeneity measurement - can we find some methylation datasets from tumors of known mixing? fraction of mixed methylation be used as heterogeneity marker (lit search) - improve this with mappability filtering?
• many libraries from same samples -> filtering reduces CV of methylation between samples? All Cs, low mappability Cs, clinically relevant low map Cs.

Description of low mappability regions length distribution, distance between them etc.

combine Figure 4 and 6 into a schematic new figure 4 Mixture of methylation

Figure 5 and 7 => supplemental

Update Figure 8 -> supplemental Figure 9 -> supplemental

Add Chaithanya and Louise to acknowledgments