Add gedi/price test data: chr19+chr22 Ribo-seq cohort

[pinin4fjords]

Summary

Test fixture set for nf-core/modules#11693 (gedi/price) under data/genomics/homo_sapiens/riboseq_expression/price/.

PRICE's Bayesian model needs a much higher candidate-ORF count than the existing chr20 fixtures provide for its expectation-maximisation to converge. This PR adds a minimised chr19+chr22 cohort that gives PRICE enough signal to call ORFs in CI.

PRICE output on this fixture set: 380 ORF rows (377-line orfs.tsv including header and metadata).

Files

File	Size	Notes
Homo_sapiens.GRCh38_chr19_22.pc_exon_masked.fa.gz	3.43 MB	chr19+chr22 primary assembly, hard-masked outside protein-coding exons
Homo_sapiens.GRCh38.111_chr19_22.pc.gtf.gz	1.66 MB	chr19+chr22 protein-coding-gene-only, lean attributes
4x bams/SRX1178088{5,6,7,8}.chr19_22.ds50.bam	1.06-1.52 MB each	4-sample Ribo-seq cohort, 50% downsample, filtered to protein-coding gene loci
4x .bai	~90 KB each
README.md	< 4 KB	derivation recipe

Each file is < 4 MB. Total: 10.88 MB across 11 files (down from 19.68 MB across 14 files in an earlier revision; see force-push history).

Why these sizes — empirical justification

These fixtures are at the floor of what makes PRICE produce non-empty output. Every dimension was minimised iteratively against a real gedi -e Price invocation. Probes that didn't make it:

Variant	Result
chr20 alone, 2-sample cohort (existing fixtures)	~235 candidate ORFs → NoiseModel.computeMeanSpline crashes during inference
chr20 alone, 6-sample cohort	reaches inference, produces 0 ORFs
chr19 alone, 6-sample @ds50	0 ORFs (PRICE estimates model but finds nothing)
chr22 alone, 6-sample @ds50	0 ORFs
chr19+chr22, 3-sample @ds50	Index 0 out of bounds for length 0 in PriceOrfInference
chr19+chr22, 2-sample @ds50	same crash
chr19+chr22 4-sample, CDS-only FASTA mask	PRICE crashes because GTF UTRs reference N positions in the masked FASTA
chr19+chr22 4-sample, ds30	0 ORFs
chr19+chr22 4-sample, ds50 (this PR)	380 ORFs ✓

The empirical floor along each axis:

• 2 chromosomes are needed. Neither chr19 nor chr22 alone produces ORFs at any downsample level (insufficient candidate ORFs per chromosome).
• 4 samples are needed. 3-sample and 2-sample cohorts crash PRICE's noise-model inference.
• Protein-coding-exon masking, not CDS-only. UTR regions in the GTF must reference real sequence (not Ns) or PRICE crashes.
• 50% downsample is the floor. 30% downsample produces 0 ORFs.

Could we go smaller?

Three avenues considered and rejected:

1. One chromosome with deeper coverage. Both chr19 and chr22 alone failed at any cohort/downsample tried; PRICE needs the candidate-ORF count from two chromosomes.
2. More aggressive FASTA masking (CDS-only). Crashes PRICE because UTRs in the GTF dereference to N positions.
3. Smaller cohort (≤3 samples). Crashes PRICE's noise model.

The 3.43 MB FASTA is the largest file in the set. It's chr19+chr22 protein-coding exons of GRCh38 primary assembly with everything outside those regions hard-masked; this is the most aggressive masking PRICE tolerates while still producing ORFs.

Source

Original BAMs: SRX1178088{5,6,7,8}.Aligned.sortedByCoord.out.bam from a full-scale nf-core/riboseq test_full run. Same upstream sample identifiers as the existing chr20 fixtures in this branch.

Test plan

• modules/nf-core/gedi/price/tests/main.nf.test in nf-core/modules#11693 consumes these fixtures via raw.githubusercontent.com URLs pinned to this PR's branch; URLs to be updated to modules after this PR merges
• nf-core modules test --profile docker gedi/price produces 380-ORF orfs.tsv with two-pass snapshot stability

[luisas]

Nice! LGTM!