spatial-velocity

spatial-velocity

When to use

Estimate velocity from measured spliced/unspliced count layers. Use spatial-trajectory when these layers are absent, or sc-velocity for non-spatial data.

Use from a step

from skills._sdk.notebook import load_skill
library = load_skill("spatial-velocity")
library.velocity(adata, method='stochastic')

Run examples/example_step.py with the step runner for a synthetic, executable example.

API

velocity(adata, *, method: str='stochastic', cluster_key: str='leiden', method_params: dict | None=None, random_state: int=0)

Estimate velocity in place from spliced/unspliced count layers.

Preprocessing filters genes, normalizes X and count layers, and builds PCA/neighbors/moments. Optional confidence, pseudotime and latent-time failures are reported in run_info warnings. VELOVI GPU results can vary between runs even with a fixed seed. scVelo velocity pseudotime uses an eigensolver without a seed argument; its results vary between runs.

:param adata: AnnData containing measured spliced and unspliced layers. :param method: stochastic (CLI default), deterministic, dynamical or velovi. :param cluster_key: Annotation used in output summaries; default leiden. :param method_params: CLI options with underscores; None keeps defaults, including velocity_min_shared_counts=30 and velocity_n_pcs=30. See references/parameters.md for method-specific controls. :param random_state: PCA/neighbor and VELOVI training seed, default 0. :returns: The same AnnData with velocity layers, graph and JSON diagnostics. :raises ValueError: Missing layers or invalid method. :raises ImportError: A backend is unavailable; use install_skill_deps.

run_info(adata, *, keep: bool=True) -> dict

Read the last velocity method's diagnostics and metric tables.

:param adata: AnnData returned by velocity. :param keep: True retains diagnostics; False removes them for CLI serialization. :returns: A dictionary with effective method controls and warnings, or empty dict.