RNAVelocity

RNAVelocity is a computational framework and software package designed to estimate RNA velocity from single-cell RNA sequencing data. It provides a unified approach to infer the directional dynamics of gene expression by analyzing the balance between spliced and unspliced transcripts, enabling the reconstruction of cellular trajectories and lineage relationships in complex tissues.

The package implements both deterministic and probabilistic models of transcriptional dynamics and is designed to work with standard single-cell data formats. It supports inputs such as loom, AnnData (h5ad), and sparse matrix representations, and is compatible with common preprocessing pipelines. The core outputs are velocity vectors for every gene in each cell, which can be visualized as velocity fields over dimensionality-reduced embeddings like UMAP or t-SNE.

Key features include data preprocessing tools for quality control, normalization, and gene filtering; versatile velocity estimation methods with smoothing options (k-nearest neighbors or kernel-based) to reduce measurement noise; handling of dropouts and uncertainty quantification; options for latent time estimation and lightweight batch effect considerations; and integrated visualization utilities for inflow and outflow patterns across the embedding.

Typical workflows start with preparing spliced and unspliced transcript counts, followed by running RNAVelocity to compute gene-wise velocities, generating a velocity embedding, and proceeding to downstream analyses such as lineage inference, pseudotime ordering, or velocity-based clustering. The framework emphasizes transparent parameterization and provides documentation and tutorials to assist interpretation of velocity vectors and their biological implications.

Limitations include reliance on the assumption that spliced/unspliced ratios reflect transcriptional dynamics and sensitivity to data quality and preprocessing choices. RNAVelocity remains a methodological option within the broader field of single-cell transcriptomics, often used alongside complementary analyses of cellular states and processes.

See also: RNA velocity, velocyto, scVelo, single-cell transcriptomics.