Strand bias on estimation of DNA methylation level

[hui_shi]

Hi Guys,

I was just wondering how you might deal with strand bias in DNA methylation calling using RRBS (reduced representation bisulphite sequencing). I know that strand bias is often an indication of erroneous alignment in NGS, so I am just not quite sure if I should filter out regions with disproportional number of plus and minus strands in estimating DNA methylation levels at CpG sites.

Thanks,

Hui

[raphael123]

I like this question ... And I am interested !

[dpryan]

It depends on how bad it is. I've seen cases where the bias is more in read#2 versus read#1, which isn't surprising. Regardless, the methylation extractor that I wrote for bison can exclude biased regions according to both read (1 vs. 2) and strand. I would expect at least some of the other methylation extractors (at least those packaged with programs that will make the methylation bias graphs as well) to have similar functionality.

[raphael123]

It depends on how bad it is. I've seen cases where the bias is more in read#2 versus read#1, which isn't surprising.

I guess you talk about pair end sequencing, we are talking about a atrand bias, a difference of methylation between FORWARD and REVERSE strand ...

[dpryan]

I guess you talk about pair end sequencing, we are talking about a atrand bias, a difference of methylation between FORWARD and REVERSE strand ...

I understand that. I mentioned that due to it being the biggest source of bias. I've only ever seen very small differences due to strand, but, as I mentioned, programs like the bison_methylation_extractor can trivially deal with cases of strand-bias in methylation. Presumably at least some others can as well.