omicverse-single-cell-sctour-trajectory

To compute pseudotime and vector fields from raw-count AnnData, use a trainer-based trajectory inference method that processes raw UMI counts with a negative-binomial loss to generate latent embeddings and vector-field outputs.

scTour trajectory inference is used for single-cell developmental pseudotime estimation, latent dynamics analysis, and vector-field visualization when datasets require trainer-based latent dynamics rather than graph diffusion methods.

Yes, trajectory inference with scTour requires raw UMI counts present in adata.X because the wrapper uses a negative-binomial loss path to accurately compute pseudotime and latent-space embeddings.

When graph-based methods are not appropriate, the best way to estimate latent dynamics is using trainer-based trajectory inference that computes pseudotime, mixed latent embeddings, and vector fields directly from raw counts.

scTour pseudotime analysis writes a pseudotime column to adata.obs, mixed latent embeddings to adata.obsm['X_TNODE'], and a vector field to adata.obsm['X_VF'] for downstream analysis and plotting.

Single-cell trajectory inference fails if the sctour package is unavailable because the wrapper depends on this external package to execute the trainer-based inference and generate the required pseudotime and vector-field outputs.