Skip to content

TEST_DATA.md

Latest commit

 

History

History
117 lines (93 loc) · 5.78 KB

TEST_DATA.md

File metadata and controls

117 lines (93 loc) · 5.78 KB
Raw

Bundled Sid test data

Vaxrank packages 1,148 selected alignment records in 58 cohorts in vaxrank/data/sid-test-data.zip. All Sid read fixtures are acquired through osteosarc 0.9.0 from the public CC0 Sid dataset. The archive contains no whole-source BAM, BAM index, dataset snapshot, or unselected regional reads. Its BAM indexes describe only the tiny selected BAMs.

The selection is explicit:

  • 49 native-coordinate retrieval cases cover the 44 original vaccine loci. Each retains one template, including its already-selected mate/alternative alignments. NTF3 retains a template carrying the compound AG>GT evidence.
  • Five reconstruction/ranking cohorts and two context cohorts retain their reviewed records. Their counts, competing alignments, low-quality reads and ambiguity are part of the regressions, so these are not resampled.
  • Two fusion cases retain seven original SAM records and their reviewed RNA windows/reference hypotheses.

These are selected test cohorts, not coverage, VAF or expression estimates. Mates and supplementary records are explicitly listed; this does not claim all mates or all alignments of every template in the original BAM were recovered.

Small Ensembl reference subsets, documented biological expectations, cached NetMHCpan outputs, and report expectations are included as separate supporting inputs. Osteosarc supplies the Sid acquisition/variant identities; it does not supply the independent Ensembl annotations or MHC prediction expectations.

Offline use

from vaxrank.sid_test_data import sid_test_data, sid_reads, sid_variants

root = sid_test_data()  # verifies and opens installed package resources
cases = root / "osteosarc/shared-v1/manifest.json"
variants = sid_variants(["MAP2-chr2-209694768"])
# sid_reads("osteosarc/shared-v1/<case>.bam") returns osteosarc.ReadSubset.

The tests use this same installed-package loader. It writes only a temporary process directory, needs no persistent cache, and performs no downloads. The existing CLI exports the 49 retrieval cases offline:

python -m vaxrank.download_test_data --output /tmp/sid-retrieval-tests --offline
python -m vaxrank.download_test_data --output /tmp/sid-retrieval-tests --verify-only

Existing output is verified, never replaced. Explicit custom --manifest downloads retain the generic shared OpenVax cache API for existing callers. They do not participate in the bundled Sid tests.

Regenerate the package subset

Install SAMtools 1.21 or later, then create a separate generator environment from the repository root (the recipe and generator are also included in the sdist):

python -m venv /tmp/sid-generator
/tmp/sid-generator/bin/python -m pip install -r examples/osteosarc_test_data/requirements.txt
/tmp/sid-generator/bin/python examples/osteosarc_test_data/build.py \
  --cache /tmp/sid-acquisition --output /tmp/sid-test-data.zip

The generator and Vaxrank's runtime both use Osteosarc 0.9. The generated bundle must pass the consumer's full test suite after replacing vaxrank/data/sid-test-data.zip. Compare the new archive's selected-record digests and supporting inputs with the existing bundle before copying it into that path; the generator refuses to overwrite an existing output.

This works with an empty acquisition cache. The checked-in small catalogue contains the selected native osteosarc.File and Variant identities from snapshot e4224eeedc18b9a0e66d9e57afcb5f9d613ed76a775e58cd56d320937cd6cf6f, plus pinned index receipts. It does not require the full historical metadata snapshot. osteosarc.extract_reads retrieves indexed regions and verifies source/index identity; the generator then retains only the records listed in recipe/selection.json.gz. A minimal set of one-base retrieval anchors covers the selected alignment spans, including explicitly required off-locus mates or supplementary records. The retrieval_regions helper derives this point cover when preparing a recipe, avoiding a separate remote seek for every selected read. The temporary regional results remain in the acquisition cache and never enter the package.

The allowlist stores alignment digests, order and multiplicity, not read bases. Native BAM digests include float-tag bits lost by SAM text formatting. Historical SAM/fusion inputs use their original text representation. Missing, changed or extra identical selected records fail generation. A freshly acquired source cannot silently redefine a regression baseline.

Repeat against the same verified cache without network access:

/tmp/sid-generator/bin/python examples/osteosarc_test_data/build.py \
  --cache /tmp/sid-acquisition --offline --output /tmp/sid-test-data-offline.zip

The read selection is deterministic. Acquisition receipts preserve real source identity, software versions and retrieval evidence; archive bytes can differ between cache locations/tool versions even when all selected records agree. bundle.json verifies every packaged file, and provenance.json records the recipe hashes and full osteosarc extraction lineage.

To update the source catalogue deliberately, first review/update the selector recipe and create the named osteosarc metadata snapshot, then run:

python examples/osteosarc_test_data/pin_catalog.py --cache /path/to/snapshot-cache

The published-allele baseline is explicit (corrections=False). In particular, MAP2's old deletion remains the historical selection baseline; the separate corrected-complex-allele regression runs against the same selected original reads. This migration does not change either biological expectation.

After regeneration, run ./lint.sh and ./test.sh. The tests check every selected record, native mitochondrial/GRCh37 retrieval, reconstruction and ranking, bundle integrity, and the actual wheel/sdist payload.