omicverse-single-cell-trajectory-inference

To compute pseudotime for single-cell AnnData, you can use diffusion_map, slingshot, or palantir algorithms via a shared wrapper, producing dpt_pseudotime, slingshot_pseudotime, or palantir_pseudotime alongside fate probabilities and PAGA summaries.

Trajectory inference on AnnData requires completed QC, normalization, PCA or embedding generation, neighbor graph construction, and clustering. You must have existing cluster labels in anndata.obs, an obsm representation, and a plotting basis to generate pseudotime and lineage summaries.

Yes, you can compare DPT, Slingshot, and Palantir trajectory analyses on the same preprocessed single-cell dataset by running each method through a shared TrajInfer wrapper to obtain corresponding pseudotime values and associated outputs for direct comparison.

Running Palantir trajectory inference requires optional external dependencies including pcurvepy2 for slingshot and mellon for computing Palantir gene trends. Fate probabilities and branch masks are computed when these dependencies are present.

PAGA topology summarizes single-cell cluster relationships by producing cluster-level connectivity summaries and visualizations after pseudotime computation. It interprets lineage structure by mapping transitions between annotated clusters in the AnnData object.

Diffusion map computes dpt_pseudotime using diffusion processes on the neighbor graph, while slingshot generates slingshot_pseudotime through fitting minimum spanning trees to clusters. Both are accessed via a shared wrapper to produce lineage summaries on AnnData objects.