rare cancer mutation detection pipeline

[dietmar13]

hello,

what is the best analysis pipeline for rare (<1%) cancer mutation detection from targeted (bait hybridization enrichment) resequencing (illumina)?

GATK?

and who has the best bait development service?
has someone experiences with MYbaits from http://www.mycroarray.com/index.html?

[GenoMax]

Not a "pipeline" as such but Illumina has a couple of variant callers (they are open source but no support is provided) that you may want to check out.

For germline mutations ISAAC variant caller: https://github.com/sequencing/isaac_variant_caller

For somatic mutations "Strelka": https://github.com/genome-vendor/strelka

[ymc]

Not a "pipeline" as such but Illumina has a couple of variant callers (they are open source but no support is provided) that you may want to check out.

For germline mutations ISAAC variant caller: https://github.com/sequencing/isaac_variant_caller

For somatic mutations "Strelka": https://github.com/genome-vendor/strelka

For germline, why would you use this ISAAC caller as oppose to the typical GATK?

[GenoMax]

For germline, why would you use this ISAAC caller as oppose to the typical GATK?

dietmar13 already had mentioned GATK in original post. ISAAC caller happens to be very fast for single samples.

[Bukowski]

what is the best analysis pipeline for rare (<1%) cancer mutation detection from targeted (bait hybridization enrichment) resequencing (illumina)?

GATK?

I think this is the worst application you could attempt to use GATK for.

MuTect on the other hand might be more useful. I presume this is what Foundation Medicine use for their hybridisation (SureSelect) based panel from the reading of the 'competing financial interests' section of this paper:

http://www.nature.com/nature/journal/v486/n7403/full/nature11154.html?WT.ec_id=NATURE-20120621#contrib-auth

VarScan2 comes out pretty well in this comparison:

http://genomemedicine.com/content/5/10/91

I'm just more interested how you determine things that are present at <1% aren't just noise?

[dietmar13]

thank you all for your suggestions

@bukowski,

we look at liquid biopsies from cancer patients and want identify tumor DNA among a large number of non-cancer cells...

I will look into MuTect and VarScan2 first...

best,

dietmar

[ymc]

@bukowski,

we look at liquid biopsies from cancer patients and want identify tumor DNA among a large number of non-cancer cells...

I will look into MuTect and VarScan2 first...

best,

dietmar

I think there should be specialized software for circulating tumor dna. I am looking for it too.

[id0]

I'm just more interested how you determine things that are present at <1% aren't just noise?

That's an excellent point. Once you get to that level, there is a lot of noise.

From my experience with different somatic variant callers, it's very difficult to get decent calls at below 5%. Often when they say "low frequency", they mean at least 5%.

I had some non-paired samples with coverage above 10,000X and even at that level, it was very difficult to get any standard variant caller to detect anything below 1%. I found LoFreq (http://sourceforge.net/p/lofreq/wiki/Home/) to be a tool that got the job done fairly well. At least it's designed specifically for ultra low frequency mutations. Even then, I was looking for known mutations, so false positives weren't much of an issue.

[Simon Anders]

First: I hope you have sequenced deeply, because, obviously, to see variants with less than 1%, you need coverage in the hundreds or better thousands, for both the tumour and the control.

If so, deepSNV might be what you need:

M Gerstung et al.: Reliable detection of subclonal single-nucleotide variants in tumour cell populations
Nature Communications (2011) 3: 811 doi:10.1038/ncomms1814

[SNPsaurus]

We carry out paired-end sequencing of short fragments, so the two reads confirm each others basecalls. With that, we can detect mutations at 1 in 5000 level. It still takes deep sequencing, but just to sample the complex population of DNA sufficiently to detect the rare allele, not (as with single-end sequencing) to try to overcome the base level of sequencing error.

[id0]

We carry out paired-end sequencing of short fragments, so the two reads confirm each others basecalls. With that, we can detect mutations at 1 in 5000 level. It still takes deep sequencing, but just to sample the complex population of DNA sufficiently to detect the rare allele, not (as with single-end sequencing) to try to overcome the base level of sequencing error.

Is that custom code or are there tools available to match variants on read pairs?

[SNPsaurus]

Any overlapper program will do the brunt of the work. There is quite a bit of custom code to take those results for this kind of purpose. Some other groups are doing the same thing more or less... http://www.biomedcentral.com/1471-2164/14/96

[woodydon]

I am sorry, but I want to ask the question:"Why do you want to detect extremely low abundant mutations?"

[dietmar13]

"Why do you want to detect extremely low abundant mutations?"

because i have a mixture of many non-cancer cells with (probably) only few cancer cells and want detect the latter using targeted resequencing...

[woodydon]

because i have a mixture of many non-cancer cells with (probably) only few cancer cells and want detect the latter using targeted resequencing...

Thanks for the reply.

Do you see only very rare mutations? Or you see other relatively more abundant mutations that are unlikely to be germline mutations. If there are more abundant somatic mutation candidates, how do you consider them?

Woody