t-SNE

Purpose

t-SNE is a nonlinear dimensionality reduction method used to visualize cells in two dimensions.

In Seurat, t-SNE is usually run with RunTSNE().

t-SNE is a visualization method. It does not create clusters by itself. Cluster labels usually come from FindClusters().

Where It Fits

Typical position:

RunPCA() -> FindNeighbors() -> FindClusters() -> RunTSNE()

After Harmony integration:

RunPCA() -> RunHarmony() -> FindNeighbors() -> FindClusters() -> RunTSNE()

Run t-SNE after choosing the reduction and dimensions used for downstream analysis.

Choose Reduction And Dims

Use the same reduction and dimensions used for graph construction.

Common choices:

Workflow	reduction	dims
Classic PCA	"pca"	selected PCs, often 1:30
SCT PCA	"pca"	selected PCs, often 1:40
CCA/SCT integration	"pca"	PCs computed from integrated assay
Harmony integration	"harmony"	selected Harmony dimensions

Run t-SNE

For classic, SCT, or integrated assay PCA:

dims <- 1:30

seu <- Seurat::RunTSNE(
  object = seu,
  reduction = "pca",
  dims = dims,
  dim.embed = 2,
  seed.use = 42,
  verbose = TRUE
)

For Harmony:

dims <- 1:30

seu <- Seurat::RunTSNE(
  object = seu,
  reduction = "harmony",
  dims = dims,
  dim.embed = 2,
  seed.use = 42,
  verbose = TRUE
)

RunTSNE() stores the result as a reduction, usually named tsne.

dim.embed = 2 sets the output embedding dimension. The default 2D t-SNE layout is usually used for visualization.

Check t-SNE

Check reductions:

Seurat::Reductions(seu)

Extract t-SNE coordinates:

tsne_embeddings <- Seurat::Embeddings(
  object = seu,
  reduction = "tsne"
)

head(tsne_embeddings)

Plot t-SNE

Plot clusters on t-SNE:

Seurat::DimPlot(
  object = seu,
  reduction = "tsne",
  group.by = "seurat_clusters",
  label = TRUE, # show cluster labels
  repel = TRUE # avoid label overlap
)

Plot metadata on t-SNE:

Seurat::DimPlot(
  object = seu,
  reduction = "tsne",
  group.by = "sample"
)

Plot gene expression on t-SNE:

Seurat::FeaturePlot(
  object = seu,
  features = "MS4A1",
  reduction = "tsne"
)

Note

t-SNE is useful for visualization, but distances between far-apart groups should not be overinterpreted.

Changing reduction, dims, or seed.use can change the t-SNE layout.