Report Sid SV and frameshift neoORFs from DNA events to RNA support

[iskandr]

The selected Sid RNA subsets now have a complete DNA → RNA → ORF report: every one of the nine SV geometries and 32 indel/splice nominations has an event sheet with the original DNA call/allele, available DNA support, transcript/event anatomy, every ORF or coding-window sequence and support by RNA product. Entries with unknown DNA origins, unresolved splice outcomes, no coding window or an unassessable allele remain explicit.

The focused search adds the remaining indel/splice nominations and FOXO3/ATP5MG joins to #433's vaccine/control subsets without downloading whole libraries. Thirty-one literal alleles were screened across four products (124 completed screens); MUC3A's nonliteral duplication remains unassessable in all four. The combined SV scope is nine geometries in 60 oriented views, including corrected dense CNN2/IGKC controls. Durable input, annotation, code/settings and sequence-quality witness pins are retained.

The strongest leads are TPST1–CRCP's altered upstream ORF, SPAG1/TECPR1/PIP5K1A frameshift windows and exploratory FOXO3–STRADA/CCDC47 ORFs; KTN1 has a short shifted tail with a stop. Original DNA VCF calls are linked for five SV geometries. The report distinguishes their caller support from RNA allele support and exact complete-window witnesses, and preserves RNA placements that differ from the nominated DNA cuts. It does not infer full mature transcripts, initiation, translation, tumor specificity or antigen presentation from these counts.

The complete appendix exports all 1,302 source-specific hypotheses as support TSV and 1,222 event-specific sequence groups as protein FASTA: 385 SV groups, 114 frameshift-window groups and 723 in-frame control groups. All 3,118 SV ORF occurrences retain strand-aware genomic blocks, exact RNA junctions, start/frame uncertainty and provenance in the compressed ledger. Alternative starts, overlapping windows and geometry aliases are not pooled or counted as independent neoORFs. The annotated-proteome comparison covers 245,535 pinned Ensembl 115 entries, using exact and I/L-collapsed 8-mers; it is not a matched-normal or normal-noncoding-ORF comparison.

Read the report, all event sheets and downloadable sequences, and the acquisition/reproduction details.

Validation: lint and all 5,761 tests pass locally (97% coverage). Independent export checks reconcile every frozen sequence/support row, all 385 SV nucleotide translations, all 3,118 ORF placements, all 41 event sheets, all 60 SV views and 124 literal + four unassessable outcomes; file hashes match. Nine selected frameshift windows retain independent original-SAM sequence/quality witnesses. Regression coverage includes VCF anchoring, both transcript/genomic strands, exon/intron boundaries, DNA-only call provenance/alias deduplication, original path junction indices and preservation of the nucleotide witness that earned a Q20 count. The Biopython audit dependency is declared in the data extra. The full run also exposed host-dependent mocked-acquisition tests (#450); their fixture now controls free space, with separate low-space and exact-8-GiB boundary regressions. The production acquisition guard is unchanged. Final-head CI then identified the undeclared SAMtools CLI prerequisite (#451); CI now installs SAMtools for the real original-BAM checks, with the requirement documented.

Version: 1.45.1, deployed to PyPI. Final-head GitHub CI passed on Python 3.9, 3.10 and 3.11. Clean-master ./deploy.sh passed lint and all 5,761 tests, built and verified both PyPI artifacts, and pushed v1.45.1 at merge commit 53e12a19c346f7826fb3e61e9e1c4d3e88042d49.

Fixes #446, #448, #449, #450, #451. Builds on merged #433 and the released dense support engine (#439/#442/#444). Portable replay remains tracked in #440; the complete 1,493-geometry catalogue and general neoORF discovery are outside this selected search, not negative results.

[coveralls]

coverage: 96.549%. remained the same — feature/sid-neoorf-read-subsets into master