Evaluating alternative solutions

[Kim Wong]

Hello

In your paper you describe how you used manual inspection of alternative solutions to cellularity and ploidy. In my results, I do suspect that a for a few samples one of the alternative solutions would make more sense based on what I know about the cancer type (eg: the optimal solution is ploidy=7 but this is very rare in my tumour type), but I am not sure how to justify this based on what the *alternative_pfit.pdf plots and *CP_contours.pdf plots are showing. I did see some examples in the supplementary materials, but I am still unsure of what to look for, eg: outliers, other patterns? Any tips would be appreciated!

Thanks!

Kim

[Francesco Favero]

Hi Kim,

I’m sorry for the delay! As many other tools, sequenza is merely fitting the segmentation results to a model. There are several reasons why the results might not be correct, some are biological reasons (eg, the presence of many subclones in the tumor samples, with very different genomes) or technical reasons (eg bad segmentation and noisy data).

If you see that the raw data are “weaving” it could be a bias due to the exome samples (normal and tumor have different capture efficency), this is unfortunately very common and it results in a lot of segments, with a lot of variation in the depth ratio. The algorithm is not capable to distinguish a “bad” segment in that way, so the results will give a high-ploidy solution.

You should look the raw data (genome_view, the last plot, or chromosome_view) if the CNV correspond to a B-allele frequency change to spot wrong segments. I’m working on a solution pf this bias in exomes (there are workaround already), and also I’m looking for improvement in the segmentation (looking for a better approach).

Generally I don’t consider the very-high ploidy solution true, unless the raw profile looks very clean, and carefully inspect the evidence that rule out lesses ploidy states.

thanks!

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.
For more options, visit https://groups.google.com/d/optout.

[Kim Wong]

Thanks, Francesco. Yes, I do see some 'weaving' in those samples with predicted high ploidy. When I select an alternative solution with ploidy around 2, some of these improve a lot.

I also see this in a few with ploidy=3. I have attached an example. Do you think I get get any accurate/useful information from this? I suspect that my data is noisy for these few samples.

Also you said below:

You should look the raw data (genome_view, the last plot, or chromosome_view) if the CNV correspond to a B-allele frequency change to spot wrong segments.

By this do you mean that if in the genome_view plot page 1, the B-allele (blue lines) is the same as the copy number in the second plot (page 2), then it is a bad segment?

Thanks for  your help!

Kim