omicverse-single-cell-sctour-trajectory

What problem does it solve?

Provide a focused, trainer-based trajectory inference path that computes scTour pseudotime, latent-space embeddings, and vector-field outputs from raw-count single-cell AnnData when graph-based methods are not appropriate or the sctour backend is required.

Core Features & Use Cases

• Produces notebook-aligned outputs: a pseudotime column in adata.obs, mixed latent embeddings in adata.obsm['X_TNODE'], and a vector field in adata.obsm['X_VF'].
• Ensures the wrapper uses a negative-binomial loss path by requiring raw UMI counts in adata.X and documents constraints when the external sctour package is unavailable.
• Use cases include developmental pseudotime estimation, latent dynamics analysis, and vector-field visualization for datasets that need trainer-based latent dynamics rather than graph diffusion methods.

Quick Start

Run sctour on my raw-count AnnData to produce sctour_pseudotime, X_TNODE, and X_VF and write them into adata.obs and adata.obsm.