RNA-seq for Variant calling

[runnerBio88]

Hi

I've read a few number of papers about this topic such:

http://www.sciencedirect.com/science/article/pii/S0002929713003832

http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0058815

I'm surprised about the results, for example, when comparing coding sites in WES, RNA-seq only finds about 33% of SNP. When compared with WGS the percentage rises up to 45%, but still quite low (I think).

The specificity of these methods are not bad, the sensitivity is not so good (depends of coverage).

I'm surprised (I'm newbie) by what I think these are poor results, I would expect that at least for expressed regions the % of SNP found to be higher. I'm also surprised by the fact that most papers conform with coverages of 10x or even 3x to try to call for a variant when by being RNA-seq coverage souldn't be big deal.

I understand that some aligners can work with splice junctions, so this should not be a problem for variant calling with RNA-seq data.

I don't know if anyone can give me some clues or some more info about this. I'm just wondering about this.

Any paper where same samples are compared by using WGS, WES and RNA-seq?
Thanks

[vivek_]

Have you looked into things like allele specific expression which could have an effect on number of common SNPs found between DNA and RNA Seq? I'm not sure if that alone could contribute to the entire divergence but its worth investigating.

[HESmith]

RNA-Seq is less sensitive for variant detection because transcript levels vary widely, whereas WGS and (for the most part) WES produce relatively even coverage. Some fraction of genes will not be expressed in your sample (and therefore undetectable), while others expressed at such low levels that the read depth is insufficient. For example, to obtain 10X coverage of a 1kbp transcript present at one-in-a-million copies would require 200M 50bp reads. Most RNA-Seq datasets are an order of magnitude smaller.

Note that there are methods for normalizing RNA-Seq libraries, but the vast majority of experiments are designed to detect differences in transcript levels (which would be obviated by normalization).

[InDel]

Generally speaking, it is preferred using WES (whole exome seq) to identify mutations and even better if you have matched patient samples with integration of Whole Exome seq and RNA-seq to increases mutation detection performance. Detecting mutations from RNA-Seq is not a typical approach to detect mutations, mainly due to the intrinsic complexity in the transcriptome (e.g., splicing, high/low gene expression level).