Abstract
Single-cell RNA sequencing enables unprecedented insights into cellular heterogeneity and lineage dynamics. RNA velocity, by modeling the temporal relationship between spliced and unspliced transcripts, extends this capability to predict future transcriptional states and uncover the directionality of cellular transitions. Since the introduction of foundational frameworks such as Velocyto and scVelo, an expanding array of computational tools has emerged, each based on distinct biophysical assumptions and modeling paradigms. To provide a structured overview of this rapidly evolving field, we categorize RNA velocity models into three classes: steady-state methods, trajectory methods, and state extrapolation methods, according to their underlying approaches to transcriptional kinetics inference. For each category, we systematically analyze both the overarching principles and the individual methods, comparing their assumptions, kinetic models, and computational strategies and assessing their respective strengths and limitations. To demonstrate the biological utility of these tools, we summarize representative applications of RNA velocity across developmental biology and diseased microenvironments. We further introduce emerging extensions of RNA velocity methods that go beyond classical splicing kinetics. Finally, we discuss existing limitations regarding model assumptions, preprocessing procedures, and velocity visualization and offer practical recommendations for model selection and application. This review offers a comprehensive guide to the RNA velocity landscape, supporting its effective implementation in dynamic transcriptomic research.