bayes refit

[yanmam...@gmail.com]

Hello Francesco,

When I ran Sequenza, sometimes it turned out the best SLPP solution is not necessary the optimal solution ( you stated in the paper already). Very often using IQRs of fitted copy numbers (from the method making genome view plot) to rank the solutions will produce the results more aligned with manual inspection.

To refine this technique further, I wonder if I can refit Sequenza bayes approach with copy number IQRs (a matrix annotated with cellularity 01.-1 x ploidy 1-7) I derived from the first round of fitting. If I can reset the priors with this knowledge, SLPP may will turn out to be more prefect. Also it may help to expose solutions more aligned with better copy number fitting.

My question is how/where to reset this matrix as bayes priors.

Thanks!

Yan

[Francesco Favero]

Hi Yan,

Thanks for the insightful comment.

Unfortunately the way it was designed the fitting, it allows to set the priors only for the copy number state, and it’s used at the segments level;
you can define a very simple histogram like data.frame to weighting the integer copy number solution (this greatly helps balancing the solution toward diploids rather then overfit aneuploidy solutions)
in sequenza.fit (the main function, but it’s the same in baf.fit) you could set something like
priors.table = data.frame(CN = (1, 2, 3), value = (1, 3, 2))
to set a priors for specific copy numbers (this would results in the priors on the CNt for each segments 1 = 1/6, 2 = 3/6, 3 = 2/6)

I’m in the process (a very slow process) of changing some of sequenza behaviour, but I haven’t thought of changing the fitting, as is doing an overall OK job (provide the data is fine).
However I’m open to suggestions, maybe I can add an optional refitting method to run between the first fit and the results step.

Could you share some of the code you are using the IQRs to refine the solution?
Maybe if I see the cellularity ploidy matrix you are talking about (is is the same as the results of sequenza.fit?) I can have a better idea of a possible implementation.

Best

Francesco

> --
> You received this message because you are subscribed to the Google Groups "Sequenza User Group" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to sequenza-user-g...@googlegroups.com.
> To view this discussion on the web visit https://groups.google.com/d/msgid/sequenza-user-group/35e42a7a-0bc5-42af-ade0-12b8af7f090dn%40googlegroups.com.