Sort of a noob here: looking for somatic mutation in tumors (NGS sequencing)

[lethalfang]

As sort of a pilot study, we used Solid 5500 (Agilent SureSelect Exome) to sequence one
tumor tissue and one matched normal tissue from the same patient.

I used the vendor-provided LifeScope Genome Analysis Software to align the sequence and did the SNP callings. The initial SNP call has about 50,000 entries.

For every SNP called in tumor, I aligned the Consensus Call of the same base position in the matched normal tissue, and then filtered out ones where the same base position in matched normal showed evidence of SNP.

I also filtered out those in the dbSNP.

We tested a small number of those potentially tumor-specific SNPs in Sanger sequencing, picking out ones with high coverage. 90% of them turned out to be sequencing error. For the other 10%, the same variant is also found in the matched normal.

Well obviously, my noob approach didn't work.
I picked out some SNPs to test because the bases had high coverage (although the reference vs. novel call is hardly 50-50, for most it's more like 80-20), although after the fact I do notice that many of them had pretty low Quality Score (QS = 20 or less).

So anyway, not too many discussions about LifeScope, do you guys actually recommend some different software for such analysis? Or is there something I'm totally missing here?

Thanks in advance.

[ymc]

I heard that the coverage for tumor should be at least 2x the normal cells, e.g. 100x vs 50x. Is this how u did it?

why don't u give the standard bwa-GATK pipeline a chance also?

[adameur]

Hi lethalfang

I've been using Lifescope quite a lot for exome analyses and usually I detect around 50k-60k SNPs with 90% or so overlapping with dbSNP. So that seems to be about the same numbers as you find..

I would be cautious to filter all dbSNP variants from the candidate list. As I understand dbSNP since can contain basically anything (including mutations). I usually use the dbSNPCommon instead for such filterings.

Also, my experience is the variants with very high coverage can be problematic (could be due to repeats or similar) so maybe they are not the ideal candidates..

[lethalfang]

Actually our average coverage was about 75X for both tumor and matched normal tissues.

About 95% of the initial SNP calls overlap with dbSNP. I know there could be meaningful mutations in the dbSNP. I filtered it anyway because I thought there might be a greater proportion of meaningful mutations not in the dbSNP. But if I can't find anything in the other 5%, I'll go back to dig into those in the dbSNP.

So adameur, what's your general experience with LifeScope?

I also have a colleague who recommended using SomaticSniper.
Has anyone used it? If so, how effective?

Thanks.

[adameur]

I'm not always impressed with Lifescope, but I think it works pretty well for variant calling.
Around 50k SNPs with 95% overlapping with dbSNP sounds good to me..

[lethalfang]

I'm trying this thing call Somatic Sniper now. I'll see how it goes:

http://gmt.genome.wustl.edu/somatic-sniper/1.0.2/index.html

[ymc]

How do you filter SNPs reported by SomaticSniper? By default it reported too many SNPs for me (9572). The same dataset with Mutect gave me 42 PASS SNPs.

[lethalfang]

How do you filter SNPs reported by SomaticSniper? By default it reported too many SNPs for me (9572). The same dataset with Mutect gave me 42 PASS SNPs.

You can filter by quality score.
Anyway, I've also used Strelka, and I found that to be easier, faster, and better than MuTect (for our data anyway).

https://sites.google.com/site/strelkasomaticvariantcaller/

[mikesh]

Generally speaking, if there is a sequencing error you cannot get rid of it if you'll sequence deeper. And 75x seems enough.
Some control, i.e. known mutation in tumor is desirable to fine-tune your pipeline. If you don't have one you can try intersect your SNPs with COSMIC database and check all the matches.