feat(galaxy): add tool wrappers for all five subcommands

[Paururo]

Five Galaxy tools, one per subcommand, ready to submit to galaxyproject/tools-iuc.

Where this sits

Three of the four prerequisites were already met, which is why this is mostly wrapper work:

Bioconda	1.0.2 published, 577 downloads
BioContainer	built automatically: quay.io/biocontainers/pathotypr:1.0.2--hef77e2b_0
bio.tools	registered as pathotypr
Galaxy wrapper	this

Nothing is vendored; the wrappers declare the conda package.

Verified

planemo lint	clean on all five, no warnings, no errors
planemo test	8 of 8 pass
Mutation check	breaking the output prefix so it no longer matches from_work_dir fails both classify tests

That last line is the one that matters: it is how we know the assertions read the real outputs rather than passing because nothing was checked.

Test data: 64 KB, and not filler

A 2 kb synthetic reference, eight markers, two samples, paired reads at about 18x, and a twelve-genome training set in two classes. Generated deterministically by a script kept next to it.

The generator asserts that every marker k-mer occurs exactly once in the reference, in both orientations. A marker matching in two places would make the tests pass for the wrong reason.

The expected calls follow from how the samples were built, and they cross-check each other:

• sample1 carries L2 to L2.1 to L2.1.1 plus a RIF marker, so classify calls L2 and reports rpoB / S450L
• split-fastq on the same sample's reads produces the same summary as classify on its assembly, which is exactly the property the marker format promises
• predict assigns the two held-out queries to L2 and L3
• match picks ref_A, which is sample1's own genome

Three things that had to be worked around

None of these is visible in the CLI help. All three were found by running the tools.

1. classify and split-fastq take the sample name from the input filename. Galaxy datasets arrive as dataset_NNN.dat, so the outputs would have been named after a Galaxy id. Inputs are symlinked to stable names first.
2. match writes the absolute paths of its inputs into the first column of its report. Those differ between instances, so the same fix applies.
3. --min-alt-percent is parsed as an integer. A Galaxy float parameter renders 95.0, which the tool rejects with invalid digit found in string. It is declared as an integer here.

That third one is worth a second look outside this pull request: min_alt_percent: u32 means a threshold of 99.5% cannot be expressed, even though the comparison itself is done in f32. For separating clonal samples from mixtures, the step from 99 to 100 is a coarse one. Nothing here depends on changing it, and the documentation does not claim otherwise, so it is recorded rather than acted on.

Not included, on purpose

Marker panels and models come from the history. A data manager plus a .loc table would let an administrator install the published MTBC panels once so users pick them from a dropdown, which is how Galaxy normally handles reference data. Shipping a data table with nothing to populate it would move the burden to administrators without helping anyone, so it is left as its own piece of work.

Submitting

The layout matches tools-iuc, so submission is cp -r galaxy/tools/pathotypr <tools-iuc>/tools/, then a pull request there. Their CI runs the same lint and tests on Linux, where the Bioconda package resolves.

Worth noting for that pull request: TB-Profiler is already in Galaxy, but fastlin is not, so there is currently no alignment-free MTBC lineage typer on the platform.

[Paururo]

Verified against the real Bioconda package

The one gap in the earlier verification is now closed. planemo test was re-run with real dependency resolution instead of a local binary:

planemo test --conda_dependency_resolution --conda_auto_install tools/pathotypr/

All 9 tests passed.

Provenance of the binary the jobs actually ran, read from the environment planemo built:

Environment	__pathotypr@1.0.2 (Galaxy's own naming)
Channel	https://conda.anaconda.org/bioconda
Build	1.0.2 h125a5a0_0, subdir osx-64
Binary	Mach-O 64-bit executable x86_64, reports pathotypr 1.0.2

The job log shows that environment being activated, and contains zero references to the local build directory, so nothing fell back to a binary that happened to be on PATH.

This exercises the whole chain the IUC CI will exercise: the <requirement> declaration, resolution from Bioconda, environment creation, execution, and the assertions. The only remaining difference is the platform, osx-64 under Rosetta here against linux-64 natively there.