mRNA BS-seq pipeline and NSUN2

In breif

• We optimized the experimental protocol to improve the conversion rates in mRNA BS-seq.
• We analyzed the pattern of artifacts in BS-seq and located the source of false positives.
• We introduced Gini-index and signal-nosie ratio into analysis to estimate the status of library conversion.
• We introduced C-cutoff and gene-specific conversion rates into our filter to remove false positives.
• We set up a new pipeline with high flexibility, robustness and parallelization level for mRNA BS-seq.
• We validated that NSUN2 is one of the major m5C methyltransferase acting on mRNA.
• We profile mRNA landscapes in multiple human/mouse tissues.
• mRNA m5C might relate to translation control.

Where are the noise?

It’s clear now that the noise in BS-seq come from many places:

1. The human-introduced Cs in hexamer.
2. The regions hard to be converted, especially some highly structured rRNA regions.
3. The false alignments and bases with bad sequencing qualities.

Solutions

The pipeline I designed have considered most of the situations:

1. Hexamer can be removed by trimming the bases at the end of reads.
2. We should use gene-specific conversion rates rather than a global conversion rate measured by spike-ins or the whole libray. Only analyze annotated regions.
3. Estimate the conversion status by Gini-index and signal-nosie ratio, and determine the C-cutoff that remove reads have more than a certain number of reads.
4. Optimize the alignment step to reduce false alignments. Normally, that should not be a matter.
5. SNPs can only destory m5C, but cannot create m5C in BS-seq.
6. Statistical tests have little effect on our analysis if we use a filter with high coverage and level.

In my hands, I guess we can let the sites with 5% methylation go, but we normally use a 10% cutoff for the biological importance of a site. BS-seq is very sensitive and reliable if you work with it correctly.

References

Back