Experimental deconvolution utility

[pawel-czyz]

This work-in-progress PR aims at sketching the deconvolution utility. The idea is to have the MVP (minimal viable product), rather than full-featured version of the tool.

What does consist of MVP?

• Small demo data, which can be used to play with the tool.
• covvfit deconv CLI utility.
• Allow for missing mutations data.
• Retrodict (and predict) mutation proportion for different loci (together with residuals) in plots.
• Retrodict (and predict) variant abundance proportions.

[gordonkoehn]

Hi Pawel, a curious observer here – are you implementing a streamlined version of Lollipop deconvolution here? If so interested to learn more.

[pawel-czyz]

It's supposed to be an entry point for this notebook. Not exactly the same scope as Lollipop (which can deconvolve multiple samples or just a single one), but rather a model which can smooth the deconvolution using selection dynamics model employed by Covvfit. It's thanks to @dr-david's vision 🙂
We don't have a document yet with clearly described maths (even though it's in the code), but in the meantime I can tell you more on a whiteboard if you are interested 🙂

[gordonkoehn]

Interesting. Yes - Interested, looking forward to next time we have a whiteboard nearby. Eager to keep in touch with latest analytics you two come up with.