Changes in version 0.99.2                        

Maintenance

  - Add the maintainer's verified ORCID to the package author metadata.
  - Replace blanket warning suppression with explicit handling of
    numeric labels, undefined correlations and unreachable graph paths.
    Preserve meaningful warnings from model fitting and Harmony
    backends.
  - Extract workflow and clustering stages into internal helpers,
    retaining the public interfaces and processing metadata.
  - Wrap vignette prose and align continuation lines consistently while
    preserving the rendered Markdown structure.

                       Changes in version 0.99.1                        

Fixes

  - Resolve the legacy Harmony matrix function only when exported,
    avoiding missing-export warnings with current Harmony versions.
  - Use scrapper::modelGeneVariances() for optional feature selection,
    ranking genes by residual variance. The updated trend fitter can
    change the selected features relative to the previous scran
    implementation.
  - Apply the existing MNN cosine normalization explicitly before
    fastMNN(), avoiding its deprecated normalization dependency while
    preserving the normalization, feature subset and caller-supplied
    backend settings.
  - Compute per-cell QC directly with the existing sparse-aware sums,
    avoiding the newly deprecated scuttle::perCellQCMetrics() call while
    preserving the metrics for single-cell and multi-cell objects.
  - Use exact PCA for tiny matrices or large requested fractions of the
    singular values, avoiding the current irlba small-matrix error and
    high-rank warnings. Retain sparse iterative PCA for low-rank large
    data.
  - Use R's default package citation until an associated publication is
    available, removing the custom citation without a publication DOI.

                       Changes in version 0.99.0                        

New

  - RegisterExistingIntegrationEmbeddings() registers latent spaces
    already stored in an object, with optional copying and automatic
    name detection.
  - RegisterIntegrationEmbedding() protects existing reductions from
    changes unless overwrite = TRUE is supplied.
  - BenchmarkIntegration() discovers unregistered integrated_*
    reductions and accepts method and reduction names without regard to
    case.
  - First submission to Bioconductor.
  - CreateBacObject() builds a SingleCellExperiment from a count matrix,
    a 10x Genomics directory, or a Seurat object, and MergeBacObjects()
    combines several such objects while keeping cell barcodes unique.
  - RunBacQC() and FilterBacCells() compute and apply
    prokaryote-specific quality-control metrics, including ribosomal RNA
    fraction.
  - IntegrateBacData(), RegisterIntegrationEmbedding(),
    RunIntegratedUMAP() and RunIntegratedClustering() provide batch
    integration and downstream embedding, with BenchmarkIntegration()
    scoring the corrected embeddings using scIB-inspired metrics.
  - aggregate_pseudobulk(), normalize_pseudobulk() and the run_edger_*()
    functions provide pseudobulk differential expression built on edgeR,
    covering both pairwise contrasts and spline-based time courses.
  - run_tata() implements Time Aware Trajectory Analysis (TATA), a
    time-informed trajectory abstraction for perturbation time courses,
    together with branch probability, gene trend, plotting and
    benchmarking helpers.
  - The packaged sce1 dataset provides a small bacterial single-cell
    example used throughout the vignettes.