Steven noticed that the M-Bias plots generated by Bismark from these files was a little wonky and asked that I try trimming them a bit more. The files were originally quality/adaptor trimmed with TrimGalore! on 20180516.
I decided to use some trimming software called fastp, since it had a lot of options (including trimming without the need for quality filtering/trimming, as well as being multi-threaded).
Looking at the M-Bias plots and the original FastQC assessment, I opted to trim the first 20bp from the 5’ end of all reads.
I followed this up with FastQC and MultiQC.