NEWS

monocle 2.6.1

knn-based density peak clustering is not general for all datasets. Rolled back to the previous densityPeak clustering algorithm and set it to be the default algorithm. A new Louvain clustering algorithm for dealing with large datasets (> 50 k cells) is added.
A few bug fixes for importCDS, exportCDS functions.

Various plot utilities for visualizing complex developmental trajectory (plot_complex_cell_trajectory), multi-way kinetic curve and multi-way heatmap (plot_multiple_branches_pseudotime, plot_multiple_branches_heatmap).
ClusterCells now in principle supports clustering for 100 k+ cells which depends on a procedure of kNN based density peak clustering algorithm. densityClust package version 0.3 are required for users to use this functionality.
New functions supporting importing scater or Seurat cds into monocle or vice versa (importCDS, exportCDS).

The default expressionFamily is now negbinomial.size, instead of Tobit. If you are using TPM or FPKM data, we urge you to convert it to relative transcript counts with relative2abs and use the negative binomial distribution in your CellDataSet objects.
Revamped clusterCells functionality based on t-SNE and densityPeak
New procedure for selected ordering genes called "dpFeature". See vignette for details.

Fixed an problem where reduceDimension would return different results on repeated runs given the same inputs.The problem was actually in DDRTree in two places: kmeans and irlba. We now call irlba with deterministically initialized eigenvectors and kmeans with deterministically selected rows of the input.
Fixed a problem in classifyCells related to joining factors and levels. This would generate annoying warnings.
Fixed the check for valid sizeFactors prior to differentialGeneTest. Without this check, differentialGeneTest would report FAIL on all genes because of factors not having enough levels.

Monocle's algorithm for converting relative expression values (e.g. TPMs) into absolute transcript counts, called "Census" has been re-designed. The new version is much more accurate. Census is available through the relative2abs() function. The interface to this function has changed, and the output values will be quite different from the old version. The main change is that the version in 1.99.0 reported cDNA counts, as estimated in each cell's cDNA library. Now, relative2abs reports estimates of mRNA counts in the lysate. These are easier to compare with experiments that estimate mRNA counts via the use of spike-in controls.
A new heatmap function plot_pseudotime_heatmap() replaces the old plot_genes_heatmap().
Lots of new documentation. More to follow in upcoming releases.

Ordering cells with DDRTree no longer compresses cells at the tips of trajectories.
Pseudo counts are now applied correctly no matter what the underlying distribution used to model expression is.
Variance stabilization is applied correctly when a CellDataSet object's expressionFamily is negbinomial.size().
The gene_id field in results from calculateMarkerSpecificity is now a character instead of a factor. This was leading to indexing errors and nonsensical semi-supervised clustering and ordering results.
BEAM() and plot_branched_heatmap() now use cBind instead of cbind, fixing an issue with sparse matrices.

The first public release of the Monocle 2 series. For a summary of new features and changes, please see: http://cole-trapnell-lab.github.io/monocle-release/features/

After grappling with Bioconductor build issues in HSMMSingleCell related to VGAM updates, we have moved to a different data layout in HSMMSingleCell, which has caused some changes in the vignette.

Fixed a bug in responseMatrix() that occurs when you don't have any genes that fail VGAM fitting