OpenVax fixture adoption¶
The common APIs are introduced by PRs #22, #23, and #24. Consumer adoption requires the Osteosarc 0.2.3 release; preparing these PRs does not claim that a release exists. The scientific baseline stays separate from the mechanics migration.
Inventory and ownership¶
| Consumer and fixture family | Shared replacement | Consumer-owned content |
|---|---|---|
Isovar sid_data exact-record pool (311 historical fixtures) |
legacy_fixtures generate/pack/verify/export, original SAM order and multiplicity |
Reviewed recipe, historical source snapshot and test assertions |
| Isovar expansion: vaccine variants, mitochondrial controls, reference/alt/other/uncallable and quality stress | select_assigned_names, write_selected_names; explicit historical QName-only policy |
Independent CIGAR classification, frozen preferred-witness assignments and producer version |
| Isovar short/long SV, three fusions, empty controls | select_window_segments and indexed extraction; explicit once-only or preserve-duplicates policy |
Breakends/exon windows, observed-alignment compatibility, reconstruction/translation |
| Isovar figure comparisons, stress/performance families | Existing exact-record recipe pool, shared source acquisition | Scientific comparison, stress design, benchmark outputs |
| Topiary compact Sid, indels, regional RNA overlays, all-variant audit, shared NTF3 | regional_corpus.generate_regional_corpus, fixture_assets.fixture_paths |
Selection declaration, annotations, translation/prediction/ranking expectations |
| Vaxrank 58 exact cohorts (49 vaccine-locus retrieval cases plus context/ranking/SV) | cohort_bundle pinning, selection, header checks, nested manifests, packing and loading |
Checked-in cohort recipe, reference/prediction metadata and ranking tests |
| Vaxrank regional corpus and reporting analyses | Existing extract_reads acquisition; source-specific selection declarations |
Library grouping, independent quality/orientation audit and reports |
| Varcode historical snapshot and source VCF fixtures | Optional current adapter pin 0.2.3; existing Dataset/Cache |
Original historical export version, alleles, VCF event definitions and protein tests; no BAM dependency added to ordinary tests |
The historical adapters keep distinct reviewed transport formats, with explicit SAM-text fidelity where appropriate. There are no runtime imports of Isovar, Topiary or Vaxrank inside Osteosarc. Their compatibility scripts contain defaults and CLI plumbing; the moved acquisition/selection/packing/verification loops have one owner. Topiary's nested manifest and Vaxrank's cohort format remain supported without requiring either checkout to use a data bundle.
Common recipe conformance¶
All three read consumers accept --panel-recipe recipe.json --panel-source
rna=archive.bam --output NEW_DIRECTORY --offline through their existing builders:
python -m isovar.sid_data generate ...python -m scripts.generate_sid_fixtures ...(Topiary)python examples/osteosarc_test_data/build.py --cache CACHE ...(Vaxrank)
The shared recipe includes exact source identity, targets, required witnesses, controls/context, explicit caps and frozen evidence-producer provenance. New variants need declarations, not another BAM writer. See fixture recipes.
Run the same source/target/policy through all three actual CLIs, with Python network connections disabled, and compare full membership, reasons, source and header identities:
python -m scripts.check_fixture_consumers --isovar /path/to/isovar \
--topiary /path/to/topiary --vaxrank /path/to/vaxrank
This explicit integration check requires the consumer checkouts. Ordinary bundle use, consumer installation and tests require no sibling checkout.
Added SV research panel¶
osteosarc/data/additional_sv_recipe.json freezes SV0461, SV0402, SV0055
(ATP8B5P region), SV0175 (GABBR1 duplication), and SV0499 (IMMT region). The sdist
ships original BAM inputs, indexes, receipts, hypotheses and a regeneration command
in tests/data/additional_svs/README.md. Four source archives retain 251 records,
with 24 witness/context members. Exact record identities preserve missing
qualities, original-query intervals, direction labels and shared source scope.
These annotations retain the saved research interpretation; no regenerated
prediction becomes its own expected-result oracle.
The named SV panel also pins the existing PARD3B/CDKN2B, GABBR1/SLC29A1 and OTUD7A/FMN1 RNA event definitions. Source-call disagreements and complex-cluster relationships remain in original records. Shared reads and repeated candidate matches never imply independent molecules.
Release and distribution gates¶
Every consumer PR bumps its own version and pins Osteosarc 0.2.3 deliberately. Existing exact 0.1.x installation pins change; recorded historical data versions and source/correction/reference identities do not. Isovar's offline fixture API is shared on Python 3.9 as well; Osteosarc's Python 3.9 suite is tested explicitly. Varcode keeps the current adapter optional and the historical collector's exact old-version requirement separate.
Before merging consumer adoption, release Osteosarc 0.2.3, run each consumer's lint/test gates, and inspect real sdists/wheels for fixture membership and size. Isovar and Vaxrank's built-distribution regressions exercise original-read loading outside their checkouts. No fixtures are acquired during package installation. Consumer scientific rebaselines require separate review, even when acquisition or regeneration succeeds.