Chapter 9: Trajectory Inference & Lineage Branching (Pseudotime)

Chapter 9: Trajectory Inference & Lineage Branching (Pseudotime)

Biological differentiation, embryonic development, and immune activation are continuous biological processes, not discrete jumps between static clusters. Trajectory inference orders cells along a continuous progression termed pseudotime, measuring the distance a cell has traveled along a dynamic developmental path.

1. Trajectory Inference with Slingshot

Slingshot builds a Minimum Spanning Tree (MST) over pre-computed cluster centroids, identifies branching lineages, and fits smooth principal curves through cells using simultaneous principal curves.

library(slingshot)
library(SingleCellExperiment)

# Convert Seurat object to SingleCellExperiment
sce <- as.SingleCellExperiment(pbmc)

# Run Slingshot specifying a stem/progenitor starting cluster
sce <- slingshot(
    sce, 
    clusterLabels = 'seurat_clusters', 
    reducedDim = 'UMAP', 
    start.clus = '0' # Specify root cluster (e.g. HSCs)
)

# Extract pseudotime lineages
pseudotime_values <- slingPseudotime(sce)

# Plot trajectories overlaid on UMAP
plot(reducedDim(sce, "UMAP"), col = topo.colors(100)[cut(pseudotime_values[,1], 100)], pch = 16)
lines(SlingshotDataSet(sce), lwd = 3, col = 'black')

2. Principal Graph Learning with Monocle 3

Monocle 3 utilizes reversed graph embedding (SimplePPT) to learn an explicit principal graph spanning single-cell manifold trajectories and detects bifurcation branch points.

library(monocle3)

# Create a cell_data_set object
cds <- new_cell_data_set(counts, cell_metadata = cell_meta, gene_metadata = gene_meta)
cds <- preprocess_cds(cds, num_dim = 50)
cds <- reduce_dimension(cds, reduction_method = "UMAP")
cds <- cluster_cells(cds)

# Learn the principal trajectory graph
cds <- learn_graph(cds, use_partition = TRUE)

# Order cells in pseudotime by choosing a root node
cds <- order_cells(cds, root_cells = root_cell_barcodes)

# Plot cells colored by learned pseudotime
plot_cells(cds, color_cells_by = "pseudotime", label_cell_groups = FALSE, label_leaves = TRUE, label_branch_points = TRUE)

3. Discovering Lineage-Dynamic Genes

Once cells are ordered along a trajectory, regression models identify genes whose expression changes significantly as a function of pseudotime, revealing the sequential cascade of transcription factors driving differentiation.

# Identify genes dynamically regulated along pseudotime in Monocle 3
pr_graph_test_res <- graph_test(cds, neighbor_graph = "principal_graph", cores = 8)
dynamic_genes <- subset(pr_graph_test_res, q_value < 1e-5)

# Plot dynamic gene kinetic curves
plot_genes_in_pseudotime(cds[c("CD34", "GATA1", "MPO"), ], color_cells_by = "pseudotime")