Somatic SNP

[wanguan2000]

about somatic-sniper result.
For an example of Somatic SNP which come from somatic-sniper :
cancer: G(40 reads)-A(16 reads)
normal : G(38 reads)-A(7 reads)
I think this SNP can't be Somatic SNP?

[RockChalkJayhawk]

SomaticSniper filtering

I too have had this problem with SomaticSniper. I reccomend filtering out those call where 5% of the alternate allele can be found in the normal sample. I would be interested in Someone from Wash U could explain why this occurs though and how they recommend getting past this issue.

[ernfrid]

Yes, this is expected behavior. The algorithm looks for differences between tumor and normal. It does not explicitly require a lack of support in the normal. One of the drawbacks and perks, depending on your situation, is that sites with a number of alleles in the normal will be predicted as somatic. Since I have spent most of my time thus far working with leukemias (where there are likely to be tumor cells present in the normal sample) and we always validate our predictions, I tend to prefer this behavior. I realize this is the minority of situations though and more stringent calling may be preferred.

In general, if you would prefer to be more conservative, I would recommend that you use the newer joint calling mode with a somewhat lower probability of somatic mutation than the default. I have not tested this extensively, but I would think using -s 0.001 or -s 0.0001 might be a reasonable start. -s0.01 gives similar behavior to the old mode, and much lower mutation rates appear to be too stringent in my opinion. The paper does show some power analyses of this calling mode.

If you are extremely confident that there should be no/low reads in the normal though, then hard filtering, as RockChalkJayhawk recommends, seems a reasonable approach to me as well.