Why methylation sequencing requires much less sequencing depth than DNA-Seq?

Question

[CY]

I am pretty new to methylation sequencing. I was under the impression that, comparing to DNA-Seq, methylation sequencing such as WGBS requires more sequencing depth given that methylation level is a continuous measurement instead of binary like in DNA-Seq. However, It turns out methylation sequencing only requires depth of 10-20x, sometime even <5x. Why is this the case?

[ATpoint]

Can you provide some references for that, especially the coverage requirements (what is the analysis goal) and where did you get it from that DNA-seq is binary?

[CY]

I sense it is common concensus that methylation sequencing only requires ~10x coverage. This is probably a reference.

By binary in DNA-Seq, I mean for detection somatic mutation (mutant vs wild type). But in methylation detection, the methylation level is continuous ranging from 0% to 100%. It seems that much higher sequencing depth is needed to precisely measure such continuous value.

[ATpoint]

A CpG is either methylated or not, it is that not both strands in each cell must necessarily share the same methylation status and in a bulk of cells you have heterogeneity. The same goes for mutations. I think no DNA-seq is ever binary, maybe single-cell applications of haploid cells. In this article they recommend 30x, no?

[CY]

Sorry for the confusion. Indeed we get heterogeneity issue no matter what we are detecting. What I meant is that, for DNA-Seq, sometime we only care whether there is a somatic mutation at given locus. The sequencing depth here is only required for us to confidently distinguishing true mutant from false posive one. However in methylation detection, there is a question of degree of methylation at each locus and this 'degree of methylation' is what we are care about. Would not depth of 20~30x be insufficient to confidiently represent this 'degree of methylation'?

[Kevin Blighe]

For somatic mutations, the high read depth in DNA-seq is desired so that one can call low frequency mutations in the sample biopsy being sequenced. Such biopsies are usually a heterogeneous mixture of tumour clones.

For calling germline variants, high read depth is absolutely not required. Our work showed that the ideal read-depth is ~30; however, one can achieve high sensitivity to Sanger sequencing at just 18 read depth. I have written more about it, here: A: DP in VCF files?

As ATpoint pointed out, methylation is just binary... methylated, or not methylated.

[ATpoint]

I would check the tool that you aim to use for differential methylation analysis or methylation calling. They typically recommend a certain coverage to have enough power at a desired minimum effect size.

Answer 1

[Charles Warden]

I would typically filter for 10x covered sites / regions, prior to running post-alignment analysis.

Strictly speaking, I have worked with Targeted BS-Seq and RRBS (not WGBS), but I would have expected the coverage requirements to be higher (for example, the alignment rates for Targeted BS-Seq / RRBS are lower than Exome DNA-Seq). So, your actual sequenced reads would then need to be higher than what you actually get aligned to your genome.

If there are clear methylation changes, I would filter regions with >70% methylation in one group and <30% methylation in the other group.

However, I often need to use |Percent Methylation| > 20%, and some effects may be even less (although those situations may have additional complications). In these situations, you actually want more than 10x coverage (maybe even more than 20x coverage).