Loss of previously called clonal events after callSubclones()

[pblaney]

Hello @sdentro @jcesar101 @davidwedge

As part of an ongoing collaboration, I've been working with a large set of WGS data with complex genomes and generated Battenberg (v2.2.9) CNV profiles for all with the hg38 reference. During our QC, we noticed some significant differences in samples previous analyzed with Battenberg using the hg19 reference as well as SNP array data (that agreed with the previous hg19 Battenberg profiles).
Specifically, we noticed some clonal gains/amplifications that coincide with sample cancer pathology were lost (called neutral) as well as deletions were being interpreted as CN-LOH.

After consulting the BAF + LogR files and comparing them to the hg19 equivalent, we quickly noticed the bands were indiscriminate of one another suggesting the final CNV profiles should not differ that considerably. Further investigation, showed the non-rounded profile and the final subclonal profile were not in agreement with these missing/incorrect calls.
The non-rounded profiles contained these events [i.e. the output from fit.copy.number()] while the subclonal profile did not [the output from callSubclones()]

We also confirmed that these segments weren't identified as one of the subclonal states in the output.
Also, the missing amplification events also couldn't be explained by read depth/coverage issues as these were comparable to regions of called gain/amplification.

Here is a quick example of one case, notice chr1q, chr6p, chr7, chrX:

I have more examples with available intermediate files to show but figured better to start the discussion before attaching numbers of screenshots.
I can create a temporary place for sharing if necessary as well.

Happy to help troubleshoot further.
Best,
Patrick