Hi all,
I spotted this issue yesterday. I have not had the opportunity to characterize it very well but wanted to determine whether anyone else had seen anything similar.
Basically, I repeated a Somatic Sniper analysis with duplicates marked by Picard when I realized it had initially been run without duplicate marking. When inspecting the results I noted that a variant candidate had disappeared between the two runs despite the fact that it still had pretty high coverage after accounting for duplicates. I then repeated the analysis AGAIN using a subset of the original BAM containing just the couple of hundred bases around the variant position and duplicates marked. This time the variant was called. Wondering if this was somehow related to duplicate marking, I ran the analysis on the full BAM again but with duplicates REMOVED. Again, the variant was called. Using -Q 20 -q 20 filters for all of chromosome 14, the number of candidate variants with duplicates deleted was around 5 lower than the number with variants marked.
So in brief: depending on whether variants are marked or deleted, Somatic Sniper appears to be producing difference numbers of candidate variants.
Has anyone observed anything similar and had the time to further investigate it?