single sample somatic variant calling software

Question

[Zhenyu Zhang]

There are many softwares that can call tumor normal pairs easily. However, in a real clinical setting, we only have tumor sample, and likely on FFPE slides.

Does anyone have any suggestions one single tumor sample variant calling pipelines without normals? I know some softwares like MuTect do have single sample mode, but likely to have many junks b/c it's not designed to do so. I am looking for more complicated tools, like the one FoundationOne is using.

Answer 1

[igor]

MuTect has a "panel of normals" option (--normal_panel argument), so you can compare to a group of normal samples. This cleans up the results significantly. See the "Normal-only calling for panel of normals creation" section at https://software.broadinstitute.org/gatk/gatkdocs/org_broadinstitute_gatk_tools_walkers_cancer_m2_MuTect2.php

FoundationOne is doing something similar to that. However, it is not as sophisticated as you think. I once checked tumor and matched normal tissue for a patient who also had FoundationOne results without a matched normal. More than 90% of FoundationOne highlighted variants turned out to be not really somatic. This was a couple of years ago and they surely improved their process, but there is only so much room for improvement.

You can also check against other databases such as 1000 Genomes or ExAC for population allele frequencies.

However, calling somatic variants without a normal is not a good idea even though you can publish them in reputable journals.

[igor]

There is a new paper from Shi et al. that summarizes this well:

Cancer-only WES pipelines are unreliable (69% somatic mutations are false positive)

Answer 2

[tamsen]

Hi Shakuras,

You are welcome to try Pisces. We use it extensively for single-sample FFPE samples.

https://github.com/Illumina/Pisces https://github.com/Illumina/Pisces/wiki/Pisces-5.1.3-Design-Document

Get 5.1.3 branch or binary for now. Should have a new release posted in a few weeks.

[tamsen]

OK, that took a while. 5.1.6 is now released.

https://github.com/Illumina/Pisces https://github.com/Illumina/Pisces/wiki

Answer 3

[de.subhajyoti]

You can also try SomVarIUS. It is developed for somatic variant identification from unpaired tissue samples. The link to the publication is: http://www.ncbi.nlm.nih.gov/pubmed/26589277 and the software can be found at http://www.sjdlab.org/resources/

Answer 4

[tim.chu90]

Our lab has been working on something like this. It's still pretty new and is only works on a cluster, but if you're willing to try it we'd be happy to work with you to get it working. https://github.com/RabadanLab/TOBI Feel free to PM me if you're interested.

[igor]

Can you briefly explain how it works? The tool is well-documented, but there isn't a lot of background info (unless you want to go through the source code). It says it's using machine learning, so that assumes that certain features make a mutation more or less likely to be somatic. What are they?

Answer 5

[daniel]

Octopus is capable of tumour only somatic and germline variant calling. There is an evaluation of tumour-only calling on WGS data in the Octopus paper.

Answer 6

[Emad]

ISOWN is a powerful tool for predicting somatic mutations in cancer tissues in the absence of matching normal tissues. https://github.com/ikalatskaya/ISOWN