Variant calls with a low fraction of alt reads

[Jeremy37]

Hello all.
I have a need to call variants for specific genes when I expect the fraction of reads having the alternate base to be very low, possibly in the range of 5% or lower. Examples of where this might be useful are to detect mutations present in a small fraction of cancer cells, or in any other tissues that are mosaic for the mutation.

I have tried to change the parameters of samtools to be as lenient as possible, but I have variants that it will not call (I use -B, I adjust -d, I've tried all the bcftools options). I have also tried GATK with the most lenient settings possible and can't get it to output such variants.

Do you have suggestions for a variant caller that can do what I want? Basically I want to get every position that has at least 3 alternate reads, regardless of the number of reference reads. I may just roll my own caller by going through the old samtools pileup output, but it will probably be super inefficient. Any advice?

[rskr]

Maybe a good statistics book?

[Heisman]

There is a program called SPLINTER you may want to look into. It's designed for pooled samples (their data shows it works well for up to 500 individuals), and if you set the expected number of alleles to be, say 200, you can call variants as low as 0.5 %.

http://genome.cshlp.org/content/20/12/1711.full

[Jeremy37]

Hmm, SPLINTER seems to be targeted at something quite different. At a minimum you need positive and negative synthetic controls for a particular type of variant that you are targeting, and this is what it uses to build an error model for calling variants in your large pool of samples. I don't think I could get around this requirement. All I have are the reads/alignment for a few individual samples.
Thanks for the idea though!

[Heisman]

A positive control isn't actually required, although a negative control is. At our sequencing center Phi-X is spiked into every lane, and that can be used as a negative control if the same applies to you.

[amaer]

I am in a similar spot, Jeremy, have you figured out anything?
Did you try software that call somatic mutations like SNVmix, varScan, SomaticSniper?

[Jeremy37]

Varscan does the trick

I am in a similar spot, Jeremy, have you figured out anything?
Did you try software that call somatic mutations like SNVmix, varScan, SomaticSniper?

Varscan seems to work well for this, letting you specify whatever thresholds you want.

http://varscan.sourceforge.net/

[amaer]

Thanks Jeremy for your very prompt reply. I was thinking too that varscan might do what I wanted, but upon reading the paper of VarScan2, it seemed that you would need at least 8% of all reads to be of a certain variant allele.

[Jeremy37]

Thanks Jeremy for your very prompt reply. I was thinking too that varscan might do what I wanted, but upon reading the paper of VarScan2, it seemed that you would need at least 8% of all reads to be of a certain variant allele.

That is just a default setting, as stated in the paper.

[amaer]

Awesome, thanks!