I realized there are differences in percent methylation between allc files, Steven’s 5x destranded bed files, and original Bismark coverage files:
- see jupyter notebook: 20190925_allc_v_5xbed_v_.cov.gz.ipynb
- see Mon. Sept 23 post
These differences impact how we validate DMRs. We have been using 5x.bedgraphs and .bam files to look at differences in methylation across samples and see if we believe when a DMR is called. However, if when a DMR is called and it was based on data that was further filtered down than the .bam or 5x.bedgraph files (e.g. by MAPQ score), then we may be validating with poor quality data. And we may not be seeing differences in methylation states across samples if poor quality data is included.
Track 1 = bedgraph from bismark (not filtered for coverage or mapq) Track 2 = 5x destranded bedgraph from steven Track 3/4 = deduplicated bam from bismark
Track 5 = 5x bedgraph from allc file (filtered for MAPQ 30) Track 6 = bam file filtered for MAPQ 30 and –min-num-ch 3 and –max-mch-level 0.7
align reads to bismark deduplicate and methylation extractor
current DMR pipeline: generate allc files: - min-mapq = 30 dmr caller: - 5x coverage
Hollies pipeline: - .cov files from bismark are filtered for 1. 5x coverage 2. covered in all samples