What Coverage For Genome Re-Sequencing By Illumina ?

Question

[helene.badouin]

Hello,

I was wondering was coverage you need to do genome re-sequencing in illumina (Illumina HighSeq 2000) ?

I was told 100x, which seems high, but I read that people often seem to use a 20-30x coverage.

Moreover, is it necessary to have a higher coverage to look for intra-population selective sweeps (from individual samples), than to investigate the genomic architecture of differenciation between sister species ?

Thank you by advance for you answer.

[Biojl]

In which species do you intend to work? You know what's the quality of their genome?

[helene.badouin]

It's a phytopathogen fungi genome, with a high GC-rate, so I think we're going to re-sequence several individuals at a high coverage at first (100x). Then we'll do some sampling, to see how much we can lower the coverage for further experiments without decreasing sensitivity.

Answer 1

[Zev.Kronenberg]

Just an example of whole genome coverage:

Rather than giving you a hard number here are two articles that answer your questions.

Assessing the accuracy and power of population genetic inference from low-pass next-generation sequencing data. Crawford & Lazzro 2012.

Low-coverage sequencing: Implications for design of complex trait association studies. Li et al 2011.

Whole genome depth modeling:

Exome dist:

[helene.badouin]

Thank you for the link. The first one in particular is very relevant for my interests (non human populations with small sample sizes).

Answer 2

[Jeremy Leipzig]

Coverage should follow a Poisson distribution, so if your mean coverage is 30X, you will fall below 20X about 3.5% of the time. In theory to get 30X at 99% of locations you will need a mean of 45X coverage.

Unfortunately the genome does not respect this distribution and you will often see deserts and hotspots with thousands of reads, although this is largely a mappability issue.

[Zev.Kronenberg]

Yup, it is naughty data. I Often see that a negative binomial is a better fit.

[Jeremy Leipzig]

using the negative binomial, what mean coverage is necessary to have 99% of bases covered at 30X?

[Zev.Kronenberg]

I guess I should have been more clear: this was for exome data. I also added a plot for WG data in my original post.

Exome depth histograms often look more like:

n<-100000 hist(rpois(n,rgamma(n,2,0.0333)))

Answer 3

[Lee Katz]

With bacteria, we are aiming for something like 50x. For high quality SNPs, we aim for 100x so that even the lower-coverage bases will have good coverage.

Answer 4

[swbarnes2]

It also depends what you are looking for. For homozygous SNPs, 30x average will do pretty well. For heterozygous, or mixed SNPs, 50x is more like it.