Tumor-only variant calling

etal · 11-26-2014, 05:03 PM

I'm attempting to call SNVs in tumor samples without a paired normal sample. Some of these tumor samples can have low purity, so the allelic frequency of somatic variants is much lower than the .5 expected for germline SNPs. FreeBayes in particular gives these variants an extremely low quality score, even with the --pooled-continuous model, making it difficult to usefully filter the resulting VCF.

Does anyone know of a good SNV caller designed for single tumor samples? Foundation Medicine claims to have built one, but did not release it publicly.

Failing that: Can I fudge FreeBayes' --ploidy or --cnv-map arguments based on known tumor purity so that it will model any fully clonal somatic SNVs properly?

This workflow looks like:

Infer copy number segments from mapped reads (CNVkit).
Infer tumor purity from segments (THetA).
Compute fake tumor-cell ploidy = (2 / tumor_purity). Example: sample is 20% tumor cells; a somatic, heterozygous SNV that is fully clonal in the tumor cells would have allele frequency .1, so fake ploidy is 10.
Rescale CNV values to fake ploidy, so neutral copy number is the fake ploidy value (e.g. 10). Round to integers.
Run FreeBayes with --cnv-map = rescaled CNVs. (Also --pooled-discrete, I suppose.)

Would this allow FreeBayes to accurately call somatic variants with appropriate quality scores?

Could the same be done with GATK UnifiedGenotyper or HaplotypeCaller, or samtools?

etal · 12-05-2014, 10:49 PM

Mentioned in another thread:

Quote:

Originally Posted by SNPsaurus

You might try LoFreq for calling rarer variants:
http://sourceforge.net/projects/lofreq/
http://www.ncbi.nlm.nih.gov/pubmed/23066108

I think this is the sort of variant caller I was looking for.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Strelka: Somatic small-variant calling workflow for matched tumor-normal samples	ctsa	Bioinformatics	15	12-15-2014 02:38 AM
Reliable variant calling	Mudal	Pacific Biosciences	3	05-12-2014 01:12 AM
Calling tumor CNVs without normal as control?	ymc	Bioinformatics	2	11-06-2013 02:36 AM
variant calling	kjaja	Bioinformatics	1	11-04-2011 08:16 AM

11-26-2014, 05:03 PM	#1
etal Member Location: California Join Date: Oct 2013 Posts: 23	Tumor-only variant calling I'm attempting to call SNVs in tumor samples without a paired normal sample. Some of these tumor samples can have low purity, so the allelic frequency of somatic variants is much lower than the .5 expected for germline SNPs. FreeBayes in particular gives these variants an extremely low quality score, even with the --pooled-continuous model, making it difficult to usefully filter the resulting VCF. Does anyone know of a good SNV caller designed for single tumor samples? Foundation Medicine claims to have built one, but did not release it publicly. Failing that: Can I fudge FreeBayes' --ploidy or --cnv-map arguments based on known tumor purity so that it will model any fully clonal somatic SNVs properly? This workflow looks like: Infer copy number segments from mapped reads (CNVkit). Infer tumor purity from segments (THetA). Compute fake tumor-cell ploidy = (2 / tumor_purity). Example: sample is 20% tumor cells; a somatic, heterozygous SNV that is fully clonal in the tumor cells would have allele frequency .1, so fake ploidy is 10. Rescale CNV values to fake ploidy, so neutral copy number is the fake ploidy value (e.g. 10). Round to integers. Run FreeBayes with --cnv-map = rescaled CNVs. (Also --pooled-discrete, I suppose.) Would this allow FreeBayes to accurately call somatic variants with appropriate quality scores? Could the same be done with GATK UnifiedGenotyper or HaplotypeCaller, or samtools?