sc-velocity

name	sc-velocity
description	RNA velocity analysis for single-cell omics. Wraps scVelo to quantify spliced/unspliced kinetics, latent time, velocity graphs, and driver gene ranking with optional dynamical mode fitting.
version	0.1.0
author	OmicsClaw
license	MIT
tags	["singlecell","velocity","scvelo","dynamics","kinetics"]
metadata	{"omicsclaw":{"domain":"singlecell","emoji":"⚡","trigger_keywords":["rna velocity","latent time","spliced","velocity graph","kinetics model"]}}
source_reliability	[{"source":"https://github.com/theislab/scvelo","description":"Official scVelo repository maintained by the scverse/Theis Lab team with active releases for RNA velocity pipelines.","score":0.93,"rationale":"First-party repo with 3k+ stars, peer-reviewed methods, and recent commits (Dec 2025) tracked via the GitHub project page.\n"},{"source":"https://github.com/K-Dense-AI/claude-scientific-skills","description":"Claude Scientific Skills pack that documents standardized skill templates, metadata, and orchestration guardrails.","score":0.9,"rationale":"Centralized skill repository from K-Dense AI used across Claude Scientific Skills deployments; provides consistent SKILL.md format adopted by OmicsClaw.\n"}]

⚡ Single-Cell RNA Velocity

Quantify transcriptional dynamics by coupling spliced and unspliced matrices using scVelo's stochastic and dynamical models. This skill ingests AnnData/loom counts, performs preconditioning (normalization, moments, phase portrait checks), computes velocities, and returns latent time embeddings with driver gene evidence suitable for trajectory-aware downstream workflows.

When to Trigger

• User mentions "RNA velocity," "latent time," "dynamic gene programs," or "spliced/unspliced layers"
• Input AnnData includes .layers["spliced"] and .layers["unspliced"]
• Need to compare lineage directionality following clustering/integration steps

Supported Inputs

Format	Notes
.h5ad	Preferred; expects layers["spliced"]/["unspliced"], obs annotations, optional var kinetic priors
.loom	Auto-converted to AnnData; verifies layers/{spliced,unspliced}
Matrices	Provide --spliced/--unspliced MTX + metadata to build AnnData wrapper

Typical Workflow

1. Load + QC

   import scvelo as scv
   adata = scv.read("sample.h5ad")
   scv.pp.filter_and_normalize(adata, min_shared_counts=30, n_top_genes=4000)
   scv.pp.moments(adata, n_pcs=30, n_neighbors=30)
   

2. Velocity Computation

   scv.tl.velocity(adata, mode="dynamical")
   scv.tl.velocity_graph(adata)
   scv.tl.velocity_confidence(adata)
   

3. Latent Time & Drivers

   scv.tl.recover_dynamics(adata)
   scv.tl.latent_time(adata)
   scv.tl.rank_velocity_genes(adata, groupby="leiden")
   

4. Visualization & Export

   scv.pl.velocity_embedding_stream(adata, basis=, color=)
   adata.write_h5ad()