Use of solid biopsy copy number caller

[TBradley27]

Considering use of other copy numbers calling methods

Hello,

Thank you for the tool!

For solid tumours, two copy number callers are provided:

• QDNAseq
• ASCAT.sc

Whilst both tools have their uses, QDNAseq only produces relative copy numbers as outputs. On the other hand, whilst ASCAT.sc does provide absolute copy number as output , it does not have associated written methodology or benchmarking in terms of how it works or operates. This provides quite a big overhead for users before use, if the users wants to understand how the copy number calling works (essentially they would have to dive into the source code to work this out)

We released some time ago a fully documented and benchmarked method (rascal) - https://www.biorxiv.org/content/10.1101/2021.07.19.452658v1 - and I think this could be more appropriate for use mostly because it is method that is properly reported and benchmarked. Another option for a published and well documented method for shallow whole genome sequencing is ACE (https://academic.oup.com/bioinformatics/article/35/16/2847/5265327?login=true)

[lbeltrame]

Hello, and thanks for the suggestions!

I'm familiar with both ACE and rascal (I believe I filed an issue ages ago there). Your suggestion has certainly merit, and I agree that a properly benchmarked method is the ideal solution for these types of pipelines.

I'll outline first why we chose ASCAT.sc:

1. At the time I was looking (before SAMURAI was made) for an absolute copy number method applicable to sWGS that wouldn't require arbitrary choices when selecting the optimal purity / ploidy combination. I saw it mentioned in the supplementary methods of the Drews et al., Nature 2022 paper, and since I was trying also to find something applicable for their CIN signatures, I looked at the repository;
2. My initial tests showed a reasonable performance and a rough concordance with the data I had at hand
3. Given this, we went with it in SAMURAI and it worked reasonably well with our test datasets

Now, let's go over your suggestions. ACE for me (I used it extensively, see Pesenti, Beltrame et al., EJC 2022) is a no-go because:

• It has a fairly bad performance (I rewrote the solution finding loop for internal use because it was fairly inefficient)
• It looks more or less unmaintained
• Without "external support" like variants, selection of the optimal solutions is fairly arbitrary, especially with low-purity samples (and we recently benchmarked it with some internal data, the ploidy estimations were often overestimated)

Rascal is a different story, because I remember using it in the past and certainly had less issues than ACE (as also outlined in the preprint you linked). I think it would be a worthwhile addition to SAMURAI. I have a few practical questions because it's been such a long time that I used it that I don't remember. Not all of these are required to get it into SAMURAI, some are nice to have.

1. Does it allow extracting the absolute CN in a programmatic way? The CLI tool that was in the repo at the time had some shortcomings,
2. Does it allow outputting both the absolute (integer) and absolute (float) copy number values for sample segments?
3. Is there a way to output per bin data (nice to have)?
4. Is there a statistic one can use to evaluate the goodness of a purity / ploidy combination (like ichorCNA's loglik; nice to have)?

Thanks again for submitting your suggestion here, it's very appreciated.