The effects of incomplete bisulfite conversion upon mapping efficiency

This question has also been posted on Biostars

I have sequenced numerous multiplexed pools of BS amplicon-seq libraries derived from human samples on a MiSeq over the past few weeks. I have been utilising trim-galore and Bismark for alignment and am finding the mapping efficiency to be really low for two pools (55% & 30% respectively) both of which had a cytosine per base sequence of around 10-20% throughout the entire read in their fastqc files.

The high C content visible in the fastqc files makes me think the poor mapping efficiency is due to poor bisulfite conversion, as this would be expected to be close to zero is bisulfite conversion had actually took place.

I am new to the field so any help would be greatly appreciated.

• It may be good to familiarize with the steps behind mapping bisulfite converted reads. In this post on biostars I give a quick breakdown of the steps involved if you want to have a look (but of course there is a lot of material available on it). – dariober Jan 17 '18 at 8:56

Bisulfite conversion efficiency has no effect on the mapping rate in bismark and similar tools. The reason is that the reads are fully bisulfite converted in silico before alignment to minimize mapping bias.

I would suggest that you play around with the settings handed to bowtie2, such as using local alignment and modifying the --score-min option to allow more mismatches. Alternatively, you might try a different aligner like bwameth, which will always do local alignment.

• Thank you. --local flag in the bowtie2 docs looks promising, I'll give this a shot when I get into the office. Will Bismark parse any bowtie2 options/flags to bowtie2 (I ask as --local is not present in bisamark docs but only within bowtie2 docs)? What can I conclude about the high cytosine per base sequence in the fastqc files? I've generally found this is at 0% in previous analyses or 10% at the ends, at which point i would trim the ends. – David Ross May 24 '17 at 7:26
• If after running bismark the MBIAS plots don't look relatively flat, you can pass a filter parameter to bismark_methylation_extractor, so that e.g. you ignore the last 10bp of each read. – 719016 May 26 '17 at 13:08
• @719016: You might be correct, it's been a few years since I've looked at the bismark code, I generally try to avoid using it since it's so incredibly slow and used to give poor results. – Devon Ryan May 26 '17 at 13:08
• As Devon Ryan also proposes, you may want to try bwameth, which is a simple wrapper around bwa-mem, and does keep the soft-clipped alignments if they exist. You can then apply another script for counting the CpGs from the bam, e.g. this one: github.com/dariober/bioinformatics-cafe/tree/master/… – 719016 May 26 '17 at 13:10
• I generally recommend MethylDackel for methylation extraction and QC, but I'm biased :P – Devon Ryan May 26 '17 at 13:12