Monocle 3 has been re-engineered to analyze large, complex single-cell datasets.
The algorithms at the core of Monocle 3 are highly scalable and can handle millions of cells.
Monocle 3 adds some powerful new features that enable the analysis of organism- or embryo-scale experiments:
A better structured workflow to learn developmental trajectories.
Support for the UMAP algorithm to initialize trajectory inference.
Support for trajectories with multiple roots.
Ways to learn trajectories that have loops or points of convergence.
Algorithms that automatically partition cells to learn disjoint or parallel trajectories using ideas
from "approximate graph abstraction".
A new statistical test for genes that have trajectory-dependent expression. This replaces both the old
differentialGeneTest() function and BEAM().
Project query data set onto a reference.
Transfer annotations to query data set from reference.
Save and load Monocle objects and transformation models.
Store counts matrix on disk rather than system memory.
Mixed negative binomial distribution for fit_models.
A 3D interface to visualize trajectories and gene expression.
Most of the algorithmic details in Monocle 3 are described in
Cao & Spielmann et al.