Changelog

Source: NEWS.md

Version 0.1.2

Released 2026-03-25.

Added

• Bundled human and mouse GENCODE gene-annotation data was added for gene-level filtering workflows. The snapshot now also stores genome-context fields such as sequence name, source, coordinates, strand, gene status/source, and annotation level in addition to gene identifiers and gene types.
• Small built-in PBMC example assets were added as pbmc_small and pbmc_small_raw, sampled from the packaged pbmc1k / pbmc3k references for check-safe examples and README workflows.
• Integration and cluster-diagnostics metrics were expanded with sn_calculate_silhouette(), sn_calculate_graph_connectivity(), sn_calculate_pcr_batch(), sn_calculate_clustering_agreement(), sn_calculate_isolated_label_score(), sn_calculate_cluster_entropy(), sn_calculate_cluster_purity(), sn_identify_challenging_groups(), and the aggregate sn_assess_integration() wrapper.
• Rare-cell-aware clustering support was added through sn_detect_rare_cells() and new sn_run_cluster() parameters that can append rare-aware features such as Gini-selected genes, local HVGs, local markers, or optional CIARA-derived features before PCA and Harmony.
• Bulk RNA-seq deconvolution support was added through sn_deconvolve_bulk(), sn_store_deconvolution(), and sn_get_deconvolution_result(), covering local BayesPrism runs plus local CIBERSORTx container workflows and result import.
• pkgdown documentation is now reorganized around workflow stages rather than source files alone. New end-to-end articles cover preprocessing and QC, clustering and integration, metrics and diagnostics, annotation and pathways, composition analysis, and interpretation/reporting.
• sn_find_de() now supports pseudobulk differential expression with limma in addition to DESeq2 and edgeR.
• sn_find_de() now supports marker discovery with COSGR when the optional GitHub package is installed.
• sn_initialize_codex_project() now scaffolds user analysis repositories from the shipped inst/codex/project-template/ assets, creating a governed project layout with AGENTS.md, memory/, docs/standards/, skills/, config/, data/, scripts/, notebooks/, runs/, and results/.
• sn_initialize_project() is now a convenience wrapper over the packaged project-template initializer.
• sn_get_codex_skill_path() now exposes packaged Codex asset paths for the Codex root, package-usage skills, project template, and project-template skills.
• sn_install_codex_skill() now installs package-usage skills, project governance skills, or both from the packaged Codex asset layout.
• Signature catalog helpers were added: sn_list_signatures(), sn_add_signature(), sn_update_signature(), and sn_delete_signature().
• Stored-result discovery and retrieval helpers were added: sn_list_results(), sn_get_de_result(), sn_get_enrichment_result(), and sn_get_interpretation_result().
• The interpretation layer now supports user-supplied background context and dual output styles for either model-facing prompt bundles or human-readable summaries.

Changed

• Internal sparse-matrix workflows were optimized for better runtime and lower peak memory use. Pseudobulk DE aggregation now groups columns without materializing dense matrices, split Seurat assay layers are combined through sparse triplet assembly instead of repeated indexed writes, exact kNN fallbacks now use blockwise distance evaluation instead of constructing a full cell-by-cell distance matrix, and gene-symbol standardization now reuses bundled annotation data plus grouped row aggregation instead of external CSV reads and per-column duplicate collapsing.
• Developer-facing informational notifications are now routed through internal package logging helpers instead of a mix of ad hoc logger, message(), and cli progress calls. Progress wording is now more consistent across initialization, clustering, enrichment, and preprocessing workflows, while real warning() and stop() conditions retain their existing semantics.
• sn_enrich() now uses a single x input with automatic dispatch across gene vectors, ranked named vectors, data frames, and Seurat-stored DE results. Its gene_clusters formula now drives both grouped ORA (gene ~ cluster) and ranked GSEA (gene ~ log2fc) workflows, and multi-database requests such as database = c("H", "GOBP", "KEGG", "C2:CP:REACTOME") are supported in one call.
• sn_enrich() now aligns MSigDB arguments with msigdbr by supporting collection / subcollection, while still accepting collection strings in database such as H or C2:CP:REACTOME.
• sn_enrich() now filters returned enrichment tables by raw p-value rather than adjusted p-value and writes one result file per requested database with stable prefix / outdir naming.
• Dataset documentation is now being consolidated into a single R/data.R source, and user-facing examples are shifting from simulated matrices or network-backed PBMC downloads to the built-in small PBMC package data.
• sn_find_de() now uses a single method argument instead of separate test_use and pseudobulk_method parameters.
• sn_find_de() now uses layer consistently for Seurat v5 workflows and no longer exposes the legacy slot argument.
• sn_filter_genes() now supports annotation-aware filtering through gene_class and exact gene_type values backed by the bundled GENCODE snapshot for human and mouse.
• Signature registry maintenance now delegates add/update/delete operations to the upstream SignatuR package API instead of maintaining a parallel custom tree-editing implementation inside Shennong.
• Harmony-backed integration now targets the immunogenomics/harmony harmony2 developer branch in package metadata and CI instead of the CRAN release line.
• Integration metrics now prefer stored Seurat neighbor graphs when available and otherwise fall back to Annoy-based approximate kNN or exact distance search, keeping the default assessment path fast enough for routine use.
• sn_get_signatures() now reads from a package-owned signature snapshot built from the full SignatuR tree during development, so runtime signature retrieval is stable, tree-structured, and no longer depends on the installed SignatuR package version.
• Signature build assets now center on data/shennong_signature_catalog.rda, which is rebuilt directly from the upstream SignatuR dataset during development, replacing the opaque R/sysdata.rda storage used by the earlier snapshot prototype.
• sn_enrich() now stores enrichment results in object@misc$enrichment_results[[store_name]] when a Seurat object is supplied, aligning enrichment with the existing stored DE workflow.
• pkgdown articles, shipped Codex skill references, and NEWS.md are now treated as required deliverables for any user-facing workflow change.
• The pkgdown deployment workflow now installs the current package before publishing the site, so reference generation can load Shennong inside the temporary deployment worktree.

Fixed

• Fixed pkgdown GitHub Actions deployment when deploy_to_branch() runs in a clean runner without a preinstalled copy of Shennong.
• Fixed sn_install_shennong(channel = "github") so it no longer tries to resolve runtime-optional GitHub Suggests by default during installation.
• Removed the optional rare-cell backends FiRE, CellSIUS, and EDGE from Shennong’s supported dependency surface and sn_detect_rare_cells() interface.
• Fixed sn_plot_dot() theme handling so the optional catplot theme no longer tries to impose an additional aspect ratio on top of coord_fixed().
• Fixed Rd example line-width failures in R CMD check and declared the runtime data.tree dependency explicitly.

Version 0.1.1

Released 2026-03-19.

Added

• Automatic human/mouse species inference through sn_get_species() using hom_genes and mitochondrial naming patterns.
• A layer-aware pkgdown article covering inferred species, non-default count layers, and stored differential-expression metadata.
• Additional regression tests for species inference, DE metadata, and BPCells-backed Seurat layers.

Changed

• sn_initialize_seurat_object() now attempts species inference before deciding whether to compute species-specific QC metrics.
• sn_get_signatures() now uses package-local fallback signatures for the core categories needed by Shennong workflows when SignatuR is unavailable.
• Stored DE results now include schema and provenance metadata such as package version, timestamp, assay/layer context, and threshold settings.
• sn_plot_*() helpers now treat catplot as an optional enhancement rather than a mandatory dependency.

Fixed

• Fixed sn_remove_ambient_contamination() for BPCells-backed Seurat layers by materializing BPCells matrices before passing them to decontX.
• Fixed optional dependency handling so missing SignatuR no longer breaks core initialization and clustering paths.
• Fixed SoupX validation order so missing raw input is reported before package availability issues.
• Fixed the sn_deconvolve_bulk() example so the cibersortx dry-run path no longer fails package examples for missing credentials.
• Fixed GitHub Actions dependency installation so check and coverage jobs keep a minimal, solvable dependency set instead of trying to install unsupported GitHub-only optional backends during lockfile generation.

Version 0.1.0

Released 2026-03-18.

Added

• sn_load_data() as the primary example-data loader.
• Consolidated clustering into sn_run_cluster().
• A unified ambient contamination interface for SoupX and decontX.
• Expanded test coverage across composition, data loading, clustering, utilities, visualization, and ambient contamination.
• Missing help pages for exported functions, CI scaffolding, a richer README, and updated package metadata for current R syntax requirements.

Changed

• Retained sn_load_pbmc() as a deprecated compatibility wrapper around sn_load_data().
• Refreshed generated documentation and package metadata as part of the modernization effort.

Fixed

• Addressed multiple modernization issues in the package build, test, and documentation pipeline.