In this vignette, SCORPIUS is used to infer a trajectory through cells in artificial single-cell RNA-seq data. Note that the dataset is generated in a very naive manner and is only meant to be used for demonstration purposes, not for evaluating trajectory inference methods.
Expression values for 384 cells and 500 genes is generated as follows.
library(SCORPIUS) dataset <- generate_dataset(num_genes = 500, num_samples = 384, num_groups = 4)
The resulting dataset is a list containing a matrix named
expression and a data frame named
expression is a 384-by-500 matrix containing the expression values of all the cells and all the genes.
sample_info is a data frame with the metadata of the cells, containing only the group each cell belongs to.
In order to infer a trajectory through this data, SCORPIUS first reduces the dimensionality of the dataset.
SCORPIUS uses Torgerson multi-dimensional scaling to reduce the dataset to three dimensions. This technique attempts to place the cells in a space such that the distance between any two points in that space approximates the original distance between the two cells as well as possible.
The distance between any two samples is defined as their correlation distance, namely
1 - (cor(x, y)+1)/2.
The reduced space is constructed as follows:
expression <- dataset$expression group_name <- dataset$sample_info$group_name space <- reduce_dimensionality(expression, "spearman", ndim = 3)
The new space is a
r ncol(space) matrix, and can be visualised with or without colouring of the
different cell types.
draw_trajectory_plot(space, progression_group = group_name, contour = TRUE)
The main goal of SCORPIUS is to infer a trajectory through the cells, and orden the cells according to the inferred timeline.
SCORPIUS infers a trajectory through several intermediate steps, which are all executed as follows:
traj <- infer_trajectory(space)
The result is a list containing the final trajectory
path and the inferred timeline for each sample
The trajectory can be visualised with respect to the samples by passing it to
draw_trajectory_plot( space, progression_group = group_name, path = traj$path, contour = TRUE )
We search for genes whose expression is seems to be a function of the trajectory timeline that was inferred, as such genes might be good candidate marker genes for the dynamic process that is being investigated.
gimp <- gene_importances(expression, traj$time, num_permutations = 0, num_threads = 8) gene_sel <- gimp[1:50,] expr_sel <- expression[,gene_sel$gene]
To visualise the expression of the selected genes, use the
draw_trajectory_heatmap(expr_sel, traj$time, group_name)
Finally, these genes can also be grouped into modules as follows:
modules <- extract_modules(scale_quantile(expr_sel), traj$time, verbose = FALSE) draw_trajectory_heatmap(expr_sel, traj$time, group_name, modules)
Any scripts or data that you put into this service are public.
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.