SV interest catalogue¶

The SV interest catalogue lists 637 structural-variant nominations from the September 22, 2026 source audit, with their original calls, breakend geometry, gene annotation, expression and RNA evidence. It ships with the package and loads offline:

from osteosarc import load_panel, load_sv_interest

catalogue = load_sv_interest()          # Targets plus source pins
targets = load_panel("sv-interest-v1")  # The same targets, for fixture recipes
macrod2 = targets["SV0203"]
print(len(targets), macrod2["kind"])

The catalogue includes the displayed ESVEE/PURPLE/LINX PASS table, focused DRAGEN calls, CTAT long-read calls and focal RNA definitions. It is not an exhaustive scan of every DRAGEN call or every possible RNA junction. Membership is a nomination for investigation, not vaccine admission or proof of translation. No entry is removed for missing RNA, annotation or an ORF.

The expressed-gene-to-intergenic set includes MACROD2 (SV0203), ITGBL1 (SV0089), ZYG11B (SV0085), RERG (SV0078), PHACTR1 (SV0368), TFDP2 (SV0381), TENM1 (SV0281), GABBR1 (SV0111), IMMT (SV0499), KLF15 (SV0172) and KMT2C (SV0377). The previously investigated FOXO3, PARD3B, OTUD7A and TPST1 rearrangements and additional chr20/ATP8B5P/GABBR1 junctions are retained too.

Coordinates, aliases and evidence¶

Each target preserves its original one-based reported coordinates and original VCF alleles where available. Resolved targets provide zero-based interbase boundaries and the retained side of each breakend. Traversal orientation is not inferred from those sides: the original oriented hypotheses remain in a separate field. Single breakends and unresolved geometries use kind="unresolved" and retain their original call instead of fabricating a mate.

adjacency_group_id groups identical retained-flank geometry, including reverse traversals. It does not establish identical inserted alleles or independent mutations. Original calls retain sample-specific alleles and LINX complex-event membership. Neither aliases nor alternative ORFs should have their counts added.

Gene annotation is pinned to Ensembl 115. Exact breakends may be intergenic even when a historical caller gives them a nearby gene name. The absence of an overlapping coding model is not a reason to exclude a target.

gene_expression contains January 2025 T2 UCLA bulk RSEM measurements matched by unique stable Ensembl gene ID, preserving the original versioned ID. It does not establish expression in T0, T1 or the organoid, or identify a mutant allele. TPM is a gene/transcript abundance measure, not protein abundance (RSEM documentation).

rna_evidence keeps each product's sample, original source URL, completion status and separate split, splice-gap and deletion-gap counts. Incomplete products have missing counts, not zero. The eight completed broad regional extractions are bounded around nominated loci; they do not exhaustively test cryptic splicing away from a DNA breakpoint. Counts are scoped templates, not independent molecules or calibrated protein abundance. Exact DNA-allele checks with 20-base flanks remain separate from adjacency geometry.

Isovar supplies reconstructed sequences, frames and complete-interval support; Topiary supplies the report/ranking policy. No protein hypothesis or biological likelihood is inferred by this data-access interface. Read-orientation flags describe processed input reads; they do not establish biological RNA strand or justify automatic rejection of reverse-complement-supported candidates.

Reproducibility¶

catalogue["sources"] records original source URLs, sizes and SHA-256 pins, including the DNA VCFs, call tables, Ensembl annotation and full gene-expression file. Every API call returns a fresh object; caller changes cannot mutate the packaged snapshot. Existing sv-regressions-v1 and vaccine-loci-v1 panels keep their established contents and policies.

osteosarc --offline fixtures panel sv-interest-v1 > targets.json

Selection and bundle generation still require an explicit source-scoped fixture recipe. Merely listing a target does not create a BAM or claim that its reads have been acquired.