Which Somatic SNP Callers to Benchmark?

Question

[sichan]

Hi Everyone,

I want to benchmark somatic variant callers. The callers should:

• handle tumor/normal genomes, but it would be nice if the caller could handle exomes, targeted resequencing, etc
• be well supported and maintained (e.g. questions/answers regarding software being posted in online forums, bug fixes and new releases being released, etc)
• published in peer-reviewed journal (or at least in submission)
• output final results in VCF

Note that somatic SNPs is the main priority at the moment.

This older BioStars post has provided a very nice summary of tools available ~2 years ago: Best Software For Detection Of Somatic Mutations From Matched Tumor:Normal Ngs Data

Given my list of requirements, I've come up with these 14 tools:

• GATK
• Isaac
• LoFreq
• MutationSeq
• MuTect
• Samtools
• Shimmer
• SomaticSniper
• Strelka
• VarScan2
• FreeBayes
• SNVer
• Platypus
• HapMuC

Are there any others you would recommend?

Thanks

Answer 1

[Chris Miller]

I have no callers to add, but will mention that in general, you'll see better performance by intersecting several different callers, as opposed to looking at each caller individually. See the bottom panel on our poster here for some results from ultra deep sequencing.

Full results are in a manuscript that's in submission, but the bottom line is that they all produce poor results on variants below about 20% VAF.

[poisonAlien]

Hi Chris, Just an observation from the poster. In the main figure for scatter plot of AML31 it says, "At relapse, the subclone represented by cluster 4 was the dominant clone, though it comprised only about 1% of the cells in the original tumor". Did you mean the cluster-3 in relapse with IDH2 mutation ? Might have been a typo there.

[Chris Miller]

Yeah, that was a typo - nice catch :)

[hongen.hugo.xu]

Did it get published?Thanks.

[Chris Miller]

Yes: http://www.cell.com/cell-systems/abstract/S2405-4712(15)00113-1

Answer 2

[Len Trigg]

Some of the ones in your list are just general variant callers, which I assume you will be manually comparing calls between the tumor/normal samples to pull out candidate somatic variants.

You might also like to try Real Time Genomics somatic caller which does joint Bayesian somatic variant detection, you can adjust the somatic prior and specify an estimate of tumor cellularity if known.