methylation data, ratios of methylated vs unmethylated reads

Question

[tonja.r]

As far as I know in the methylation data they use ratio of methylated vs unmethylated reads. How do they overcome the problem of small number of reads? Let say for one specific position there are 4 methylated and 8 unmethylated reads. The ratio would be 4/8=1/2. On another position there are 40 methylated and 80 unmethalyted reads, ratio is again 1/2. However, one would trust the second position much more because the first could be an affect of an error etc.

So, how do they account for the low and high number of reads when calculating the ratios?

Answer 1

[Devon Ryan]

Using a ratio or only the percent methylation would be a terrible idea and only works with sufficiently high coverage.

In general, people should (and hopefully only ever do) use either:

1. The integer number of methylated and unmethylated observations at a position.
2. The percent methylation (in your example, 4/(4+8)=33%) and the total coverage (in your example that'd be 12).

Option 1. is better than 2. for storing things since you inevitably lose precision when you write a percentage into a text file, but that's a different discussion. The only BSseq program that I'm aware of that doesn't do what I mentioned above is RnBeads...and I can't recommend using that in general unless you have high (I would guess minimum 15x, likely more like >30x) coverage with many samples. That should be doable with RRBS, but few people have the budget for that for WGBS.

[tonja.r]

Can you please suggest some papers where they use the strategies (1 and 2) you mentioned?

[Devon Ryan]

Any paper using pretty much any of the bioconductor packages except RnBeads.