How many reads is sufficient?

[LeonDK]

Hi all,

I want to do analysis of differentially expressed genes and splice-variants.

I have Illumina HiSeq2000 101bp paired end reads mapped to GRCh38, release 78, but I am uncertain how many mapped reads are 'enough' for a sample to be included, such that I in good conscious can say, that I use 'high quality' data?

1 million? 10 million? mapped reads/sample

Cheers,
Leon

[GenoMax]

Here is a guide to get you started: https://genohub.com/next-generation-...-guide/#depth2

It would be somewhat dependent on how good your libraries are also.

[mbblack]

Generally speaking, what people are comfortable with for DGE is nowhere near what is ideal for high confidence SNP calling. Illumina has a 2010 white paper looking at depth of coverage and SNP calling that you might want to look at - http://www.illumina.com/content/dam/...sequencing.pdf.

It was from analyses like this that the idea of 30x coverage as a good rule of thumb seems to have come from. But I've seen a trend in recent literature for even higher coverage. Lately, for really high confidence calls (i.e. very low false postiive expectation) some cancer studies have been using as high as 100x coverage. However, keep in mind that for detecting rare variants, a large number of biological replicates is at least as important as depth of coverage of any given sample.

Really, it comes down to how sensitive do you need to be in your variant calls. If you need really low expectation of false positives, then you need high read depth coverage. If you need to detect very rare variants, then you need many replicates from the population under study.

P.S. you might want to read this Nature review as well - https://rtsf.natsci.msu.edu/_rtsf/as...20coverage.pdf