PRs welcome? #1

Dear Navin Lab,

While I have absolutely no experience in biology or mathematics, I am an experienced software engineer, and I thought it might be a meaningful experiment to find an avenue towards putting my spare time to lowering general friction in cancer research.

From cursory poking around, I found that CopyKAT was quite widely used. I also saw this draft rewrite and thought perhaps I could be useful assisting with some polish from the software engineering side of things. I have spent some days now familiarizing myself with the code base and prototyping, and I thought maybe I could begin by proposing some performance optimizations.

I have taken the liberty of offering a few low-risk PRs. I have limited myself to proposing only changes which reliably produce bit-identical results, as tested across hundreds of hg20 samples. I have used Claude but have audited the submissions myself. The performance improvements in each individual PR is not huge, but they add up. Output remains bit-identical. I kept the PRs focused to emphasize auditability.

Out of personal curiosity, and to better familiarize myself with this tool, I have also been experimenting with using GPU acceleration (CUDA) and have been able to achieve 2-3 orders of magnitude performance improvements compared to the R version, running the Xenium 170k dataset in 96 seconds. But the output is no longer strictly bit-identical so I don't have verification yet.

(See preliminary accuracy and performance benchmarks)

I'd also be happy to contribute in terms of workflows, versioning/releases, library packaging, automated testing etc, but I didn't want to just propose PRs for that out of the blue. If it's of interest please let me know. I'd also be happy to suggest approaches.

Looking forward to your response.