Choosing tumor type for mixed IMPACT cohort

[oliverartz]

Dear Park lab team,

Thanks for sharing SigMA - amazing tool. I have a quick question: I would like to do mutational signature deconvolution of IMPACT samples with various tumor types (36 different primary sites, some with unknown primary). How would I deal with this in run()? Would I need to loop over samples for each tumor type and adjust tumor_type as best as possible or are there any more general settings I can use to find signatures in a heterogeneous population?

If adjusting tumor_type is the solution, would it still be fair to compare the signatures between tumor types? N.b. I am not interested in Sig3, but other signatures such as MMRd, APOBEC, etc.

I would appreciate your help with this!