To ensure that Monocle was installed correctly, start a new R session and type:
library(monocle)
Getting help
Questions about Monocle 3 should be posted on our Google Group (https://groups.google.com/forum/#!forum/monocle-users). Please use
monocle.users@gmail.com (mailto:monocle.users@gmail.com) for private communications that cannot be addressed by the Monocle user
community. Please do not email technical questions to Monocle contributors directly.
The Monocle 3 workow
Before we get into the details of ordering cells along a trajectory, it's important to understand what Monocle is doing. The ordering workow shown
above has ve main steps, each of which involve a signicant machine learning task.
Step 1: Normalizing and pre-processing the data
To analyze a single-cell dataset, Monocle rst normalizes expression values to account for technical variation in RNA recovery and sequencing
depth.
Step 2: Reducing the dimensionality of the data
Next, to eliminate noise and make downstream computations more tractable, it projects each cell onto the top 50 (by default) principal components.
Then, you as the user choose whether to reduce the dimensionality further using one of two non-linear methods for dimensionality reduction: t-SNE
or UMAP. The former is an extremely popular and widely accepted technique for visualizing single-cell RNA-seq data. The latter is faster, and often
better preserves the global structure of the data but is also newer and therefore less well tested by the single-cell community. Then, Monocle 3 will
cluster your cells, organize them into trajectories, or both.
Step 3: Clustering and partitioning the cells
Monocle 3 can learn multiple disconnected or "disjoint" trajectories. This is important because many experiments will capture a community of cells
that are responding to a stimulus or undergoing differentiation, with each type of cell responding differently. Because Monocle 2 assumes that all of
your data is part of a single trajectory, in order to construct individual trajectories you would have to manually split up each group of related cell
types and stages into different sets, and then run the trajectory analysis separately on each group of cells. In contrast, Monocle 3 can detect that
some cells are part of a different process than others in the dataset, and can therefore build multiple trajectories in parallel from a single dataset.
Monocle 3 achieves this by "partitioning" the cells into "supergroups" using a method derived from "approximate graph abstraction"
(https://www.biorxiv.org/content/early/2017/10/25/208819) (AGA) (Wolf et al, 2017). Cells from different supergroups cannot be part of the
same trajectory.
Step 4: Learning the principal graph
Monocle 3 provides three different ways to organize cells into trajectories, all of which are based on the concept of "reversed graph embedding".
DDRTree is the method used in Monocle 2 to learn tree-like trajectories, and has received some important updates in Monocle 3. In particular, these
updates have massively improved the throughout of DDRTree, which can now process millions of cells in minutes. SimplePPT works similarly to
DDRTree in that it learns a tree-like trajectory, but it does not attempt to further reduce the dimensionality of the data. L1Graph is an advanced
optimization method that can learn trajectories that have loops in them (that is, trajectories that aren't trees).
Once Monocle 3 has learned a principal graph that ts within the data, each cell is projected onto the graph. Then, the user selects one or more
positions on the graph that dene the starting points of the trajetory. Monocle measures the distance from these start points to each cell, traveling
along the graph as it does so. A cell's pseudotime is simply the distance from each cell to the closest starting point on the graph.
Step 5: Differential expression analysis and visualization
Once this is complete, you can run tests for genes that are specic to each cluster, nd genes that vary over the course of a trajectory, and plot your
data in many different ways. Monocle 3 provides a suite of regression tests to nd genes that differ between clusters and over trajectories.
Monocle 3 also introduces a new test that uses the principal graph directly and can help nd genes that vary in complex ways over a trajectory with
loops and more intricate structures.
Tutorial 1: learning trajectories with Monocle 3
In this tutorial, we demonstrate how to use Monocle 3 (alpha release) to resolve multiple disjoint trajectories. We will mainly introduce:
评论1
最新资源