All notable changes will be documented in this file.

- Add
`fit_nb_offset`

to support vst.flavor=‘v2’ by default

- Updated cpp utilities to adhere to C++17 standards
(
`std::random_shuffle`

->`std::shuffle`

) - Handling of extra variables in
`latent_var`

when`vst.flavor="v2"`

- Changed
`get_nz_median2`

to support`genes`

argument; thanks @boomanaiden154 and @ScreachingFire. #155 - Replaced
`get_nz_median`

with faster alternative`get_nz_median2`

across all calls - Removed
`get_nz_median`

- Updated
`make_cell_attr`

to be flexible for named vectors; thanks @moi-taga #171

- Specify required Matrix version to >= 1.5.0

- Add
`make.sparse`

to handle`dgCMatrix`

coercsions

- Convert bitwise operators to boolean operators in utils.cpp

`vst.flavor`

argument to`vst()`

to allow for invoking running updated regularization (sctransform v2, proposed in Satija and Choudhary, 2021. See paper for details.`scale_factor`

to`correct()`

to allow for a custom library size when correcting counts

- Add future.seed = TRUE to all
`future_lapply()`

calls

- Wrap MASS::theta.ml() in suppressWarnings()

- Fix logical comparison of vectors of length one in
`diff_mean_test()`

`compare`

argument to the nonparametric differential expression test`diff_mean_test()`

to allow for multiple comparisons and various ways to specify which groups to compare- Input checking at various places in
`vst()`

and`diff_mean_test()`

- Major speed improvements for
`diff_mean_test()`

- Changed the
`labels`

argument to`group_labels`

in`diff_mean_test()`

- Fix bug where factors in cell attributes gave error when checking for NA, NaN, inf

- Ability to control the values of latent variables when calculating corrected counts
- Offset model as method, including the ability to use a single estimated theta for all genes
- Nonparametric differential expression test for sparse non-negative data

- Improve poor coefficient initialization in quasi poisson regression
- When plotting model, do not show density by default; change
bandwidth to
`bw.nrd0`

- Updates to C++ code to use sparse matrices as S4 objects
- Add check for NA, NaN, Inf values in cell attributes

- Remove biocViews from DESCRIPTION - not needed and was causing problems with deploying shiny apps
- Fix bug where a coefficient was given the wrong name when using
`glmGamPoi`

(only affected runs with a batch variable set)

- Add a
`qpoisson`

method for parameter estimation that uses fast Rcpp quasi poisson regression where possible (based on`Rfast`

package); this adds`RcppArmadillo`

dependency

- Remove
`poisson_fast`

method (replaced by`qpoisson`

) - Use
`matrixStats`

package and remove`RcppEigen`

dependency - Use quasi poisson regression where possible
- Define cell detection event as counts >= 0.01 (instead of > 0) - this only matters to people playing around with fractional counts (see issue #65)
- Internal code restructuring and improvements

- Fix inefficiency of using
`match.call()`

in`vst()`

when called via`do.call`

- Add support for
`glmGamPoi`

as method to estimate the model parameters; thanks @yuhanH for his pull request - Add option to use
`theta.mm`

or`theta.ml`

to estimate theta when`method = 'poisson'`

or`method = 'nb_fast'`

- Add a
`poisson_fast`

method for parameter estimation that uses the`speedglm`

package and`theta.mm`

by default - Add ability to plot overdispersion factor in
`plot_model_pars`

- Add and return time stamps at various steps in the
`vst`

function - Add functions to calculate grouped arithmetic and geometric mean per
row for sparse matrices (
`dgCMatrix`

) - might come in handy some time

- Default theta regularization is now based on overdispersion factor
(
`1 + m / theta`

where m is the geometric mean of the observed counts) not`log10(theta)`

; old behavior available via`theta_regularization`

parameter - Refactored model fitting code - is now more efficient when using parallel processing
- Changed how message and progress bar output is controlled; integer
`verbosity`

parameter controls all output: 0 for no output, 1 for only messages, 2 for messages and progress bars - Increased default bin size (genes being processed simultaneously) from 256 to 500
- Better input checking for cell attributes; more efficient calculation of missing ones

- Some non-regularized model parameters were not plotted

- Add function to generate data given the output of a vst run
- Add cpp support for dense integer matrices
- Minimum variance parameter added to vst function

- Rcpp versions of utility functions
- Helper functions to get corrected UMI and variance of pearson residuals for large UMI matrices

- lots of things