- Preparing for BioConductor submission.
- Slightly adjusted
waldTestGEE()
andscoreTestGEE()
to be more efficient.
- Switched GEE fitting back to use
scale.fix = FALSE
and substituted a fixed value for the Negative-binomial overdispersion parameter (instead of estimating via method-of-moments) as it improves model fits. - Added option to use a Lagrange Multiplier (Score) test for GEE mode instead of the default Wald test. The relevant argument is
gee.test
intestDynamic()
. - Updated documentation and some tests.
- Added column called
Null_Fit_Notes
to output fromgetResultsDE()
to describe when and how null models fail. This doesn't happen frequently, but it's good info to have when it does. - Expanded test suite to include C++ functions.
- Fixed incorrectly-implemented Lagrange Multiplier test after initial code was seen to be incorrect. See this paper for the relevant implementation details.
- Changed GEE fitting to use
scale.fix = TRUE
throughout the package, as it appears to be faster and more statistically efficient based on simulated data benchmarking.
- Minor bug fixes.
- More improved matrix operations using
RcppEigen
. - Improved support for
cell_data_set
objects frommonocle3
. - Sped up GLMM mode.
- Minor bug fixes.
- Improved matrix inversion through judicious usage of C++.
- Added S3 method
summary.scLANE()
to be used on output fromtestDynamic()
.
- Sped up GLMM mode.
- Switched GEE mode to use model-based variance and made computation of sandwich variance-covariance matrix conditional to speed things up and reduce memory usage.
- Sped up the NB LASSSO implementation in
fitGLMM()
. - Fixed some errors related to intercept-only
marge
models. - Added DF and KC corrections to new function
biasCorrectGEE()
, usage of which is set toFALSE
by default intestDynamic()
. - Added support for
monocle3
objects of classcell_data_set
throughout the package, notably intestDynamic()
.
- Added small-sample bias correction method to GEE sandwich variance-covariance matrix, results in smaller Wald test statistics.
- Parallelized
getResultsDE()
usingfuture
backend. - Added a function called
chooseCandidateGenes()
to identify good genes for trajectory DE testing based on mean / SD expression and sparsity.
- Added implicit regularization of selected basis functions to the GLMM mode using a NB LASSO.
- Switched candidate knot subsampling to a uniform sequence of candidate knots across pseudotime's support.
- Added
geneProgramDrivers()
function to compute & test correlations of expression with gene module scores. - Updated documentation & unit tests.
- Added
geneProgramSignificance()
function to estimate associations between gene program module scores and pseudotime.
- Added progress bar to
testDynamic()
.- Changed parallel backend in
testDynamic()
fromdoParallel
todoSNOW
in order to make this possible.
- Changed parallel backend in
- Updated documentation with more runnable examples.
- Passing
BiocCheck
with no errors. - Reduced set of exported functions to just what's necessary for model fitting & downstream analysis.
- Reverted changes made to null model scoring internally that lead to worse knot selection i.e., knot selection is improved now.
- Added DOI badge to README.
- Better compression of included datasets.
- Added
geneProgramScoring()
for module scoring of dynamic gene clusters. - Added
plotModelCoefs()
to annotate gene dynamics plots with a table of model coefficients. - Added citation file with link to Zenodo repository (until preprint is up).
- Added runnable examples to most functions.
- Added Zenodo tracking.
- Added simulated dataset to
data/
.
- Preparing for BioConductor submission i.e., reformatting code, adding documentation, etc.
- Added convolution function
npConvolve()
to be used for e.g., heatmap smoothing.
- Added the
getKnotDist()
function to pull the set of empirically-identified knots for a user-provided gene set. - Minor enhancements & documentation improvements.
- Added a function named
embedGenes()
that takes a smoothed counts matrix as input & returns PCA & UMAP embeddings along with a graph-based clustering. - Updated the
clusterGenes()
function to be much more efficient as well as changing the distance metric used to be cosine distance. - Added
theme_scLANE()
for output plots. - Enhanced documentation.
- Increased test coverage.
- Added a function named
sortGenesHeatmap()
that aids in the creation of expression cascade heatmaps by sorting genes according to where in pseudotime their peak expression is. - Changed the parameter
approx.knot
in thetestDynamic()
function to use (stochasticity-controlled) subsampling instead ofseq()
to reduce candidate knot space. - Added
summarizeModels()
to sum up slopes across pseudotime intervals.
- Changed input format of all functions to allow counts matrices formatted as
SingleCellExperiment
orSeurat
objects, sparse matrices, or dense matrices. - Updated visualization functions to reflect changes made in
ggplot2
v3.4 (mostly changing thesize
parameter in line-based geoms to belinewidth
instead).
- Added a
NEWS.md
file to track changes to the package.