search

google / deepsomatic

DeepSomatic is an analysis pipeline that uses a deep neural network to call somatic variants from tumor-normal and tumor-only sequencing data.
BSD 3-Clause "New" or "Revised" License
143 stars 15 forks source link

Feature request: support panel-of-normals (PoN) #25

Closed lbeltrame closed 1 month ago

lbeltrame commented 1 month ago

Matched tumor-normal analyses can be problematic if the normal sample has issues, is missing for any reason, or simply has an outlier genotype. A better approach, currently as far as I know only implemented in the Mutect 2 variant caller, is to use a reasonably large panel of normals (10+) to weed out also those variants appearing in a normal population (Mutect 2 flags variants present in at least two samples of the PoN).

Especially in settings where samples can come from archival sources (i.e. hospitals, or biobanks) and thus the normal samples may have not been collected, cannot be collected, or were lost for whatever reason, an approach using a PoN is definitely superior to a tumor-normal matched approach.

kishwarshafin commented 1 month ago

Hello @lbeltrame ,

I believe you are referring to the tumor-only mode of DeepSomatic? There are case studies and models for tumor-only mode and we have desribed the method in our manuscript

Screenshot 2024-09-20 at 9 37 03 AM

If you are asking about TiN contamination, it's in figure 3B. For PON of DeepSomatic, we use dbSNP, clinvar and 1KGP variants. Please let me know if you have any further questions.

kishwarshafin commented 1 month ago

Closing this issue, please feel free to reopen.