Adding leafcutter2 module

[jaempawi]

Add LeafCutter2 splicing calling module

Adds splicing_calling_leafcutter2.ipynb alongside the existing splicing_calling module, so LeafCutter1 can eventually be retired. LeafCutter2 adds a functional label to every junction — PR (productive), UP (unproductive / NMD target), NE (ambiguous), IN (intergenic) — appended as a fifth colon field on the intron ID, e.g. chr10:134786:179993:clu_1_+:PR (Dai et al., 2025).

The existing module is untouched; this one sits beside it.

Workflows

Workflow	Purpose
leafcutter2	regtools junction extraction, then clustering, classification and quantification
leafcutter2_preprocessing	the first half only, stopping at the list of .junc files
leafcutter2_cluster	entry point for junctions extracted elsewhere
star2junc	converts STAR SJ.out.tab to the BED6 junction format LeafCutter2 reads

leafcutter2 keeps a default of 1 since its junctions always come from regtools.

Guards

Each one tested to fire:

• chromosome naming disagreement between junctions and GTF, checked before clustering
• junction lists mixing BED6 and BED12, which one offset cannot suit
• empty result after filtering, rather than a later opaque ValueError
• unrecognised strand values, before they reach regtools as literals
• colliding BAM basenames, which would silently overwrite a sample's junctions

The [global] block now derives BAM paths, strandness and sample IDs from one filtered row collection, so a blank or NA path drops that sample without shifting the next one's strand. This bug is inherited verbatim from splicing_calling.ipynb, which still has it — worth a follow-up there. It is not silent corruption: SoS fails loudly with Length of provided properties (3) does not match number of groups (2).

GTF, FASTA, .fai and the junction files named in --juncfiles are now declared dependencies. --output-prefix was added so two workflows can share one --cwd, and so a manifest named cohort.v1.sample_list.txt does not collapse to cohort.

Dependency: LeafCutter2 >= 2.0

star2junc calls leafcutter2-star2junc, which exists only in 2.x.

The only conda build of leafcutter2 in any of our channels is 0.0.1 (dnachun/noarch, 2026-01-27) — conda-forge and bioconda have none. It packages the pre-2.0 repository scripts with no __init__.py and no .dist-info, and its classification step fails with:

AttributeError: module 'leafcutter2.Transcript_tools' has no attribute 'ClassifySpliceJunction'

@dnachun is building 2.0.1 for the channel. Once it lands:

1. add leafcutter2 to StatFunGen/pixi-setup envs/full_packages.txt, beside regtools
2. the skipped tests below go live, and the remaining workflows can be verified on SCC

To reproduce the verified run before then — note LeafCutter2 is not on PyPI, so pip install leafcutter2 fails:

python -m venv ~/lc2-venv
~/lc2-venv/bin/pip install git+https://github.com/leafcutter2/leafcutter2.git   # 2.0.1
export PATH=~/lc2-venv/bin:$PATH

Tests

tests/notebooks/molecular_phenotypes/calling/test_splicing_calling_leafcutter2.py, currently 2 passed, 1 skipped:

• test_cluster_requires_explicit_offset — needs no binary and no reference data, so it runs on every CI runner today
• test_star2junc — skips unless leafcutter2-star2junc is on PATH, which doubles as the >= 2.0 probe
• test_cluster_and_classify — additionally gated on XQTL_REFERENCE_DATA, since the GTF and FASTA are not in the repository; asserts the five-field ID, the PR/UP/NE/IN label, that cluster_ratios.gz cells carry /, and that junction_counts.gz does not

The guards mean CI is green now and the coverage becomes real the moment the conda package lands, without a red X in the meantime.

Files

• code/SoS/molecular_phenotypes/calling/splicing_calling_leafcutter2.ipynb
• tests/notebooks/molecular_phenotypes/calling/test_splicing_calling_leafcutter2.py
• tests/fixtures/splicing_calling/leafcutter/protocol_example.sample_list.txt
• tests/fixtures/splicing_calling/leafcutter/protocol_example.bam_list.txt
• tests/fixtures/splicing_calling/leafcutter/protocol_example.junction_files.txt
• tests/fixtures/splicing_calling/leafcutter/SAMPLE_001.SJ.out.tab
• tests/fixtures/splicing_calling/leafcutter/SAMPLE_002.SJ.out.tab

The two SJ.out.tab fixtures are new, derived from the shipped .junc files, so the star2junc MWE runs off the repository as written (~1 s). All four MWEs use the tests/fixtures data with protocol_example.* naming and SAMPLE_001/SAMPLE_002.