Strange bisulfite libraries

[frozenlyse]

Our lab is doing more and more WGBS libraries (human) - and we sporadically sequence a weird library that aligns poorly (~15% vs usual ~80% with bismark2) and has a weird base composition profile - however if we repeat the library prep from scratch it works fine.

Here's the fastqc plot.

Has anyone seen this? It appears our bisulfite conversion isn't being very efficient. We are using the Illumina protocol - except we use a homebrew bisulfite method instead of the epitect kit - we've never had problems with our homebrew method for other applications ie PCR based methylation sequencing.

Any suggestions/ideas?

[EpiBrass]

I often see this too in my libraries. I'm preparing from Epitect bisulphite kit and library prep with Epicentre EpiGnome kit. What I do is just headcrop the first 16 bases using trimmomatic. Have you tried doing this and seeing if the alignments increase? At least this would give you an idea of whether it's a problem with the whole read or just the start.

This may result in more non-unique mappings in bismark as you will have shorter read lengths. It's not so much an issue with my data as I'm using 520 bp paired end (260 bp each end) so loosing 16 bases isn't such a big thing.