What to do in a case of low mapping efficiency ?

[Emm2342]

Hi !

I have WGBS samples that I trimmed with trim_galore and mapped on its reference genome (a mitochondrial genome) using bismark (Bowtie 2). I also mapped the samples on two spike-ins : lambda (unmethylated) and pUC19 (fully methylated).

Based on the lambda spike-in, 3 samples have good bisulfite conversion efficiency (> 98%) and 5 samples have very poor conversion efficiency (as low as 70%)... The methylation calls are present on both CpGs and CpH, but the rate is higher on CpH for all 5 samples. The results for the pUC19 spike-in were good (methylation rate >97%), but again, the CpH methylation rate was kind of high, considering that methylation should only be on CpGs.

Re-doing the experiment is impossible for us. Is there any way to recover and analyse these samples while filtering the false-positives ?

I don't want to calculate DMRs, and I have for premises that 1) CpH methylation could be present in the mitochondrial genome, and 2) methylation rates/patterns could vary among samples.

Thanks !!

[FelixKrueger]

Hi @Emm2342

I am afraid it is difficult for me to give you a clear cut answer to your questions. Like with so many other experiments (e.g. spike-ins for ChIP-seq, ERCC for RNA-seq etc), using spike-ins as conversion controls in bisulfite experiments is notoriously very unpredictable. Not sure if spike-ins have ever worked consistently for anyone? I remember having had Lambda spike-in "controls" that suggested the conversion rate had been only 70%, when clearly it must have been 99% or higher (as evidenced by non-CG methylation in the genome of between 0.3 and 1%...). I think fairly often, in such cases people chose to ignore that they had included spike-ins in the first place, and continue with their analyses.

I wanted to link some thoughts about bisulfite conversion and spike-ins, and secondly a nice paper where the authors clearly showed that structural properties of mitochondria were responsible for the seemingly elevated levels of methylated mitochondrial DNA: Evidence Suggesting Absence of Mitochondrial DNA Methylation

I hope that you might find some inspiration in there?

[Emm2342]

Thank you @FelixKrueger for your quick answer.

I find that the methylation status of my samples seems to be linked to the spike-in's conversion rate. We did took the advices from Mechta et al. for our WGBS samples, so the mitochondrial DNA structure should not be the cause of these results. It could be a simple expermiental problem...

I'll have to approach the analysis very carefuly... Maybe trying to remove the problematic reads could help.

Is methylation_consistency has for premise that non-CpG should not be methylated ?

[FelixKrueger]

I am not sure whether methylation_consistency is the right script as it assesses how consistent the calls are within a read. You might want to look at filter_non_conversion (or similar), as this aims to remove reads that seemingly resisted the bisulfite conversion. I do agree though, you should approach the data analysis very carefully.