                       Changes in version 1.11.2                        

  - The decontX-package help page now has a Value section, which fixes a
    BiocCheck warning.
  - Documentation is regenerated with roxygen2 8.1.0, and the Stan model
    exports with rstantools 2.7.1. Results don't change.

                       Changes in version 1.11.1                        

  - decontX no longer depends on the celda package (ADR-0003 Stage 1).
    The one internal use of celda::normalizeCounts() (proportion
    normalization inside simulateContamination()) is replaced by the
    package's own fastNormProp routine, which produces identical
    results, and the example and vignette now plot cluster labels and
    marker-gene expression on the UMAP with scater::plotReducedDim()
    instead of celda's plotting functions. This drops celda (and its
    transitive dependencies) entirely and removes the R CMD check
    deprecation warnings that came from celda's masked, stale copy of
    decontX being invoked after library(celda) in the examples.
  - The cell cluster initialization used by decontX() when z is not
    supplied (log-normalization, variable gene selection, PCA, and UMAP)
    has been migrated from scater/scuttle to the scrapper package,
    because scuttle deprecated logNormCounts()/normalizeCounts() in
    Bioconductor 3.24. This removes the deprecation warnings from R CMD
    check.
  - Results of decontX() without z may differ slightly from previous
    versions (same workflow and parameters; different variable-gene
    criterion and UMAP implementation). A new legacyInit = TRUE argument
    reproduces the previous behavior; it requires the scater package
    (now in Suggests) and is kept for backwards compatibility for as
    long as the upstream functions remain available. Users supplying
    their own z are unaffected.
  - With the default initialization, the UMAP step itself is
    deterministic even when seed = NULL (scrapper seeds it explicitly),
    but overall decontX() results with seed = NULL remain
    non-reproducible because the EM initialization is unseeded — set a
    seed for full reproducibility.
  - The plotting functions (plotDecontXContamination(),
    plotDecontXMarkerPercentage(), plotDecontXMarkerExpression(),
    plotDensity(), plotBoxByCluster()) no longer use the deprecated
    ggplot2::aes_string() or trans = scale argument, which removes the
    ggplot2 deprecation warnings they emitted. decontX now requires
    ggplot2 (>= 3.5.0).
  - plotDecontXMarkerPercentage() and plotDecontXMarkerExpression() now
    work when x is a dense or sparse matrix (they previously failed with
    a ggplot2 aesthetics error when the plot was drawn). The facets of
    plotDecontXMarkerExpression() now follow the order of groupClusters,
    matching the x-axis.
  - decontPro() no longer draws 1000 unused samples from the variational
    posterior. Only the mean of the approximation is used, but the draws
    scaled with the number of ADTs x droplets and could take hundreds of
    GB of disk and memory on large datasets (#46). On a 50 x 400 test
    dataset the temporary file shrank from 444 MB to 2 MB and peak
    memory from 3.0 GB to 1.2 GB. Results are unchanged. rstan's "Pareto
    k diagnostic" warning is no longer shown: it is estimated from the
    draws and would always report Inf with the minimal number of draws
    now requested.

                       Changes in version 1.10.0                        

  - No significant changes (Bioconductor 3.23 release).

                        Changes in version 1.8.0                        

  - No significant changes (Bioconductor 3.22 release).

                        Changes in version 1.6.0                        

  - No significant changes (Bioconductor 3.21 release). Includes the
    DecontPro changes released in 1.4.1 on the Bioconductor 3.20 branch.

                 Changes in version 1.4.1 (2025-02-16)                  

  - decontPro() gained an ambient_counts argument, so an empty-droplet
    count matrix can be supplied to estimate the ambient contamination
    profile empirically instead of inferring it from the filtered
    droplets.
  - Breaking: the first argument of decontPro() was renamed from object
    to filtered_counts for consistency with the new ambient_counts
    argument. Code that passed the count matrix by name
    (decontPro(object = x, ...)) must be updated; positional calls are
    unaffected.
  - The estimated ambient contamination profile is now returned in the
    parameters element of the decontPro() output (p_est).
  - Updated deprecated array syntax in the Stan model, fixed a C
    compiler warning in matrixSums, and corrected a deconPro typo in the
    DecontPro vignette.

                        Changes in version 1.4.0                        

  - No significant changes (Bioconductor 3.20 release).

                        Changes in version 1.2.0                        

  - No significant changes (Bioconductor 3.19 release).

                        Changes in version 1.0.0                        

  - First Bioconductor release (Bioconductor 3.18), providing decontX()
    for ambient RNA decontamination of single-cell RNA-seq counts and
    decontPro() for single-cell protein expression (CITE-seq/TotalSeq)
    data.

                 Changes in version 0.99.5 (2023-10-19)                 

  - First submission to Bioconductor after review.
