[jaempawi]

Adds isoform_calling_torino.ipynb, which quantifies isoform usage with Torino (Buen Abad Najar et al., 2026, bioRxiv 10.64898/2026.10.21.760529). Torino factorizes each gene's read-depth matrix across samples by empirical Bayes Poisson matrix factorization into coverage components and their per-sample usage, without transcript annotations. It therefore captures intron retention and alternative polyadenylation, which junction-based LeafCutter2 does not quantify.

• Workflows: torino (coverage, per-gene depth matrices, factorization, usage tables) and torino_preprocessing (the first two steps).
• Wrappers: torino_coverage.sh, torino_count_matrix.R, torino_factorize.R and torino_usage.R in code/script/molecular_phenotypes/calling/, replacing the paper pipeline's Python and positional-argument scripts.
• Outputs: a raw usage BED, a normalized BED (as in the paper's QQNorm_factors.py) and a phenotype-group file for TensorQTL grouped permutation.
• Fixture: 24 simulated single-end BAMs over AP000547.3 (chr22), drawn from three Ensembl transcripts, with build_fixture.R.
• Tests: 8 in tests/notebooks/molecular_phenotypes/calling/test_isoform_calling_torino.py.

All 8 tests passed on SCC. On the fixture, component usage correlated with the simulated fully spliced, alternative-last-exon and intron-retention transcripts at 0.98, 0.99 and 0.99. The end-to-end test is skipped where the torino R package is not installed; its GitHub-only dependencies (torino, fastTopics, smashrgen, ebpm, vebpm, and mr.ash at a8d0b23) are not yet in the dnachun channel.