Skip to contents

Changelog

Source: NEWS.md

XCNV 0.0.0.9000

  • Selected DuckDB’s explicit non-shared home mode when supported, keeping internal overlap connections silent and README rendering deterministic across DuckDB releases.

  • Tightened ACMG/ClinGen relation labels: contigs now require non-missing, nonempty character/factor labels without surrounding whitespace. Gene biotypes reject surrounding whitespace and case-insensitive placeholder tokens, and functional_biotypes must be a nonempty unique explicit vocabulary. Legitimate explicit nonfunctional Ensembl-style biotypes remain valid and produce the contract’s 1B/3A absence criteria when appropriate.

  • Hardened the native ACMG/ClinGen scorer contract: an omitted policy_id defaults to the native policy, but a supplied value must be nonempty and exactly acmg-clingen-2019-constitutional-cnv on every row. Foreign or mixed policies now fail before a score or classification is emitted; evidence provider identity remains provenance rather than a voting or rewritten field.

  • Hardened native ACMG/ClinGen input evidence and BED geometry: required provenance/identifiers and supplied contract labels now require character or factor labels, rejecting logical, numeric, list, and matrix values before conversion; score labels are parsed without factor-code coercion. Model attribution now uses only selected rows contributing to each group or summary score, emits a scalar only for a complete common selected model, and never broadcasts an unselected model. CNV, gene, and dosage coordinates must be exactly representable integers in [0, 2^53) before overlap; strings beyond that range fail before conversion. A zero-row CNV relation is neutral, while nonempty CNVs require nonempty, provenance-labelled gene and dosage resources and therefore cannot fabricate negative evidence. Missing, empty, logical, or list-valued gene biotypes and functional-biotype vocabularies now fail before filtering, so they cannot produce 1B/3A absence evidence. Contig names remain exact caller/provider identifiers, and a valid contig with no matching resource remains a no-overlap result rather than an error.

  • Corrected ACMG/ClinGen contract resolution to be per CNV. Evidence rows now retain their source provider identity (including gene/dosage providers), while summaries use only the policy-kernel provider. Group and summary provenance is scalar only when unique, with counts otherwise; clinical classes emit only for supported status, and unrelated releases are not concatenated.

  • Added an explicit relational contract for provider, assembly, resource release, model/policy identity, and supported/unsupported/conflict status. The X-CNV compatibility lane remains GRCh37-calibrated; the ACMG/ClinGen 2019 lane validates coherent declared assemblies and keeps model scores separate from clinical classes. DECIPHER is documented only as an optional, access-controlled evidence source and is not downloaded or bundled.

  • Bundled the published 41-node X-CNV classifier as a validated runtime-neutral R data object, and made it the default when an annotation resource directory does not supply an override. Runtime prediction remains independent of XGBoost.

  • Bundled a compact real ClinVar 72461/SOX2 example with its case-local X-CNV annotations and objective GENCODE/ClinGen content evidence.

  • Corrected the published-resource readers: LJB26 START == END records are one-base points rather than empty BED intervals, LJB scores are selected by their explicit site index, and the headerless merged-sample file retains its first observation.

  • Added an installable R package API for reading CNV tables, validating resource bundles, computing X-CNV annotations, predicting MVP scores, and writing the legacy table format.

  • Added a small deterministic fixture and tinytest coverage, including an independent R overlap oracle for the production DuckDB inequality join.

  • Removed bedtools and XGBoost from the installed runtime. A deterministic staging script converts a separately obtained XGBoost model to a validated, tabular xcnv-tree-v1 artifact; the package evaluates that artifact in R.

  • Added an independently authored ACMG/ClinGen 2019 constitutional-CNV scoring contract with provenance-required evidence rows, published score ranges, criterion-group caps, and five-tier classification. A DuckDB-backed content provider derives only sections 1, 2A, and 3 from explicit gene and ClinGen dosage relations; case, phenotype, inheritance, partial-gene, and population evidence remain curator/provider inputs.

  • Recorded the pinned upstream compatibility scope and current resource limitations.