Changes in version 0.6.2 (2026-03-19)                  

26/02/2026

  - Improved error checks in deconvolute() to help with issue #20.

New features

  - Add measurement of bias and variance to tune_deconv().
  - Add plot_tunevar() to compare MSE, bias^2 and variance.
  - Add S3 method for kappa() to compute the condition number of the
    spillover matrix.
  - Add plot_path() to show how the coefficient paths vary with
    compensation for either a single bulk sample or a cell subclass.
  - Add plot_pred() as an alternative to plot_set() for viewing
    predicted vs observed results from simulations in a single
    correlation plot.
  - Add zero_fraction argument to generate_samples().

                        Changes in version 0.6.1                        

13/01/2026

  - Fix subtle division by near 0 issue in nest_output

New features

  - Massive speed up of deconvolute().
  - Add ridge parameter lambda to deconvolute() [experimental].
  - Add resvar metric to tune_deconv() to examine residual variance of
    bulk gene expression.

Important change

  - deconvolute() no longer returns se and hat in the subclass list
    element. Standard errors can be obtained with the new function se().
  - cores argument has been removed from deconvolute()

Important bugfix

  - R 4.5.2 for macOS arm64 (Apple M* Macs) has switched to using a
    faster BLAS (vecLib?) by default, which causes errors with
    parallelisation in mclapply(). The problem is isolated to R 4.5.2
    arm64 for macOS on M* Macs; or any version of R for macOS including
    intel, if vecLib BLAS is being used via a symlink. The solution is
    to use cores = 1 with deconvolute() and tune_deconv() whenever
    vecLib BLAS is in use.
  - Removed use of pbmcapply::pbmclapply() as this caused problems with
    R 4.5.2 for macOS arm64 (Apple M* Macs) even with cores=1. This
    fixes indefinite hanging in tune_deconv() associated with vecLib
    BLAS.

                 Changes in version 0.5.7 (2025-12-11)                  

11/12/2025

  - Change log argument in deconvolute() to logged_bulk. NB. this is a
    change of logic.

                        Changes in version 0.5.6                        

07/10/2025

  - Add residuals.deconv() to allow recalculation of full residuals
    matrix.

                        Changes in version 0.5.5                        

10/09/2025

  - Fix bug in mergeMarkers() if cellMarkers object has no cell grouping
    vector.

                 Changes in version 0.5.4 (2025-09-15)                  

09/09/2025

  - Fix CRAN checks
  - Switch from using cat() to message()

                        Changes in version 0.5.3                        

26/08/2025

  - Massive speed up (4-5x) in compensation optimisation

                        Changes in version 0.5.2                        

30/07/2025

  - Expanded to 3 methods for identifying outlier genes (var.e, Cook's
    distance, Studentized residuals)
  - Improved (nested) parallelisation of tune_deconv()
  - Rewrite weights code (faster)

                        Changes in version 0.5.0                        

14/07/2025

  - Calculation of SE
  - Detection and multipass removal of problematic genes with extreme
    residuals

                        Changes in version 0.2.0                        

24/01/2025

  - This is the initial build of cellGeometry