dorado is not on bioconda because of its Oxford Nanopore commercial licence. I'm using the official

nanoporetech/dorado

Docker Hub image directly (SHA-pinned to v1.4.0):

container "nanoporetech/dorado:shac8f356489fa8b44b31beba841b84d2879de2088e"

This is the same pattern used by the

nf-core/modules

parabricks modules (

<http://nvcr.io/nvidia/|nvcr.io/nvidia/>...

), which also bypass bioconda due to licensing.

conda null

is set with a comment explaining why. ONT doesn't publish semver Docker tags yet — I've opened an issue asking for them: nanoporetech/dorado#1584. Once they do, we can replace the SHA with a clean version tag. Is this approach acceptable for dorado?

2. GPU label —

process_gpu

flagged as non-standard by lint The module uses

label 'process_gpu'

because dorado requires a CUDA GPU.

nf-core/configs

defines

process_gpu

but it's not in the standard set lint checks against (

process_single

,

process_low

,

process_medium

,

process_high

,

process_long

), so lint warns about it. Is

process_gpu

the correct label for GPU-required tools? Should I also add a secondary standard label (e.g.

process_high

)?

3. CI tests — GPU not available on GitHub Actions Real basecalling tests require a GPU and can't run in CI. The module has: - 2 stub tests (run in CI — cover output file structure, versions) - 2 real GPU tests (documented, local-only) The stub tests use a tiny synthetic pod5. The real tests use a 10-read GIAB HG002 pod5 (PR nf-core/test-datasets#1968 pending merge for public hosting). Is stub-only CI acceptable for GPU tools? Are there other GPU modules in nf-core I can look at for precedent? PR branch:

sahuno:add-dorado-basecaller

Thanks!

PS: This message was written with help of coding agent