Skip to content

Trajectory — TSCAN (unsupervised)

When you do not have a supervising phenotype, TSCAN infers a trajectory from the embedding alone: cluster samples with a Gaussian mixture (BIC-selected if n_clusters=None), build a minimum spanning tree on cluster centroids, pick the longest path through the tree, and order samples along that path.

Call

from sampledisco.sample_trajectory.TSCAN import TSCAN

tscan_results = TSCAN(
    AnnData_sample=pseudo_adata,
    column="X_DR_sample",
    n_clusters=None,
    output_dir="sampledisco_demo_output/rna",
    grouping_columns=["sev.level"],
    origin=None,
    pseudotime_mode="rank",
)

TSCAN now runs on the single sample embedding uns['X_DR_sample'] — the two-key X_DR_expression / X_DR_proportion split is gone, so there is only one trajectory to compute. TSCAN returns a results dict (clusters, MST, principal path, ordering, and pseudotime); it also writes tscan_pseudotime_main and tscan_cluster columns back into AnnData_sample.obs.

Output

Writes → sampledisco_demo_output/rna/TSCAN/ ({column} is the embedding key, here X_DR_sample):

  • clusters_by_cluster_{column}.png — points colored by GMM cluster.
  • clusters_by_grouping_{column}.png — points colored by each entry in grouping_columns.
  • {column}_pseudotime.csv — per-sample pseudotime (with trajectory_type, branch_id, and cluster columns for the main path and any branches).

See the API page for the full parameter list.