Single-cell RNA sequencing (scRNA-seq) has revolutionized biology by providing a high-resolution look at the cellular composition of tissues. However, scRNA-seq suffers from a fundamental limitation: it is a destructive measurement. To sequence a cell, you must lyse it, capturing a single, static snapshot of its gene expression at that exact microsecond. It is the molecular equivalent of a still photograph of a bustling city intersection—you can see where every car and pedestrian is standing, but you cannot definitively tell which way they are walking, or how fast.
To reconstruct how cells change over time, early computational methods relied on "pseudotime" algorithms, which order cells along a trajectory based on their transcriptomic similarities. While baseline transcriptomic signatures can predict a cell line's general differentiation propensity before induction (Changbin Sun et al. in Transcriptome variations among human embryonic stem cell lines are associated with their differentiation propensity), static snapshots cannot capture the real-time, highly dynamic transitions that occur once differentiation begins. Crucially, pseudotime lacks inherent directionality; it cannot distinguish whether a cell is differentiating forward into a specialized state or dedifferentiating backward into a progenitor state.
As reviewed by Brian Hie et al. in Computational Methods for Single-Cell RNA Sequencing, a solution to this problem lies hidden within the raw sequencing reads themselves: the ratio of unspliced to spliced mRNA. This ratio provides a directional vector—an RNA velocity—that estimates the time derivative of gene expression, allowing researchers to predict a cell's short-term future state from a single static snapshot.
The Intuition of Splicing Kinetics
To understand why RNA velocity works, we must look at the life cycle of an mRNA molecule. When a gene is transcribed in the nucleus, it initially contains both coding regions (exons) and non-coding regions (introns). This molecule is unspliced pre-mRNA ($u$).
Next, the cell’s splicing machinery cuts out the introns, turning it into spliced mature mRNA ($s$), which is transported to the cytoplasm to be translated into protein. Eventually, the mature mRNA is broken down and recycled through degradation.
$$\text{Transcription } (\alpha) \longrightarrow \text{Unspliced } (u) \overset{\text{Splicing } (\beta)}{\longrightarrow \text{Spliced } (s)} \overset{\text{Degradation } (\gamma)}{\longrightarrow} \emptyset$$
This sequential process creates a predictable delay between transcription and the appearance of mature mRNA. We can exploit this delay to determine whether a gene is actively turning on or shutting off:
- Up-regulation: When a cell activates a gene, the transcription rate ($\alpha$) spikes immediately. Unspliced pre-mRNA ($u$) accumulates rapidly because the splicing machinery cannot keep up instantly. As a result, the cell exhibits an unusually high ratio of unspliced to spliced mRNA.
- Down-regulation: When a cell decides to shut a gene down, transcription drops to zero. The pool of unspliced pre-mRNA ($u$) plummets almost immediately, while the mature spliced mRNA ($s$) clears out much more slowly based on its natural degradation rate. This leaves the cell with an unusually low ratio of unspliced to spliced mRNA.
Imagine a graph where the $x$-axis represents the abundance of spliced mRNA ($s$) and the $y$-axis represents unspliced mRNA ($u$) across thousands of individual cells. If a population of cells is in a steady state—where mRNA production perfectly balances degradation—the ratio forms a predictable diagonal line.
Cells actively up-regulating the gene will bend above this diagonal line, while cells down-regulating the gene will drop below it. By calculating this displacement for thousands of genes simultaneously, we can construct a high-dimensional velocity vector for every individual cell, revealing exactly where it is headed.
Unspliced (u)
^
| / [Up-regulation: high u/s ratio]
| / .
| / .
| / .
| /____________ [Steady-State Line]
| / .
| / .
|/ . [Down-regulation: low u/s ratio]
+------------------------------------------> Spliced (s)
This kinetic framework has allowed researchers to map complex cellular trajectories across a wide array of biological systems. For example, Christian F. Guerrero‐Juarez et al. utilized splicing kinetics in Single-cell analysis reveals fibroblast heterogeneity and myeloid-derived adipocyte progenitors in murine skin wounds to discover that a subset of contractile wound-healing fibroblasts unexpectedly originates from immune lineages. Similarly, Tracy Tabib applied this directional analysis in Myofibroblast transcriptome indicates SFRP2hi fibroblast progenitors in systemic sclerosis skin to uncover a precise, two-step differentiation pathway that drives pathological skin scarring.
Overcoming the Steady-State Assumption
The classic RNA velocity model relies on a major simplifying assumption: that all genes across all cells share a constant, global splicing and degradation rate. In complex, multi-lineage systems, this assumption frequently breaks down, leading to errors in trajectory inference.
To solve this, Philipp Weiler et al. developed an expanded framework known as CellRank 2, detailed in Unified fate mapping in multiview single-cell data and CellRank 2: unified fate mapping in multiview single-cell data. Instead of relying solely on naturally occurring splicing ratios, this framework incorporates experimental pulse-chase metabolic labeling data.
By tracking chemically labeled nucleotides over precise time intervals, the method directly quantifies cell-specific transcription and degradation rates. This dynamic resolution allows researchers to distinguish between distinct regulatory strategies—such as whether a cell maintains its identity through high molecular turnover or stable, low-turnover expression.
Unlocking these precise kinetics also provides a window into how genes regulate one another. In spliceJAC : transition genes and state‐specific gene regulation from single‐cell transcriptome data, Federico Bocci et al. demonstrated that combining unspliced and spliced counts allows for the mathematical inference of state-specific gene regulatory networks (GRNs). By evaluating the stability of these networks, the authors can predict the specific "transition genes" responsible for destabilizing a cell state and driving it toward a new lineage.
Integrating Space and Attractor Landscapes
Cells do not differentiate in a vacuum; their fates are heavily steered by their physical position within a tissue and signaling from their neighbors. Modern iterations of RNA velocity have successfully fused splicing kinetics with spatial transcriptomics to map these environmental forces.
In Spatial transition tensor of single cells, Peijie Zhou introduced a multiscale dynamical model demonstrating that the concentration of unspliced mRNA can serve as a measurable proxy for the "attraction force" that a stable cellular state exerts on its surroundings. This allows for the mathematical reconstruction of non-equilibrium attractor landscapes, making it possible to distinguish between cells securely anchored within a stable phenotype and those poised on the precipice of a transition path.
Attractor State A Attractor State B
(Low Velocity) (Low Velocity)
\ / \ /
\_____/ \_____/
^ ^
\ /
\____ Poised Transition State ___/
(High Unspliced Concentration /
Strong Velocity Vector)
To handle the technical challenges of multi-batch spatial data, Tianyu Liu et al. developed spVelo in spVelo: RNA velocity inference for multi-batch spatial transcriptomics data. By pairing Variational Autoencoders (VAEs) with Graph Attention Networks (GATs), spVelo models cellular trajectories by combining intracellular splicing ratios with intercellular spatial proximity. This spatial-kinetic integration enables researchers to validate active cell-cell communication: if a signaling molecule is truly active, its presence should physically correlate with a downstream transcriptomic velocity vector in neighboring target cells.
Uncovering Disease Origins and Intercepting Fates
By transforming static transcriptomic snapshots into predictive vector fields, RNA velocity has become an invaluable tool for identifying the origins of diseases and designing targeted therapies.
Tracing Cellular Origins in Cancer
In highly heterogeneous tumors, pinpointing the exact cell-of-origin is notoriously difficult. Xiaojun Ren et al. applied velocity modeling in Single-cell transcriptomic analysis highlights origin and pathological process of human endometrioid endometrial carcinoma to trace the directional flow of gene expression across disease stages, proving that endometrial cancer originates specifically from unciliated glandular epithelial cells.
Similarly, Giorgia Maroni et al. used velocity analysis in Identification of a targetable KRAS-mutant epithelial population in non-small cell lung cancer to isolate a conserved, highly sensitive progenitor population driving lung adenocarcinomas, opening up a new avenue for early therapeutic ablation.
Overturning Developmental Dogma
Splicing kinetics can also challenge long-held assumptions about cell lineages. In Defining the marker and developmental trajectory of myeloid-derived suppressor cells in aging by single-cell transcriptomics, Yaru Su et al. investigated the accumulation of immunosuppressive cells during aging. While these cells were traditionally believed to be immature progenitors, RNA velocity vectors revealed that they actually emerge from mature neutrophils and monocytes, completely reframing how we view age-related immune decline.
Predictive Therapeutics
Instead of treating a tumor based purely on its current composition, RNA velocity allows clinicians to anticipate its next move. Chen Tang et al. designed comboSC in Personalized tumor combination therapy optimization using the single-cell transcriptome, a framework that integrates velocity data to prioritize drug combinations. By accounting for the predicted future states of cells, the system selects drugs that can intercept active cellular transitions, such as blocking a tumor cell's slide toward immune exhaustion.
This predictive paradigm has been widely adopted to map cellular hierarchies across diverse pathologies:
- Tumor Microenvironments: Tracking the directional polarization of tumor-associated macrophages (Qianqian Zhang et al. in Interrogation of the microenvironmental landscape in spinal ependymomas reveals dual functions of tumor-associated macrophages), dissecting the transcriptional control of intratumoral regulatory T cells (Feng Shan et al. in 1049 Reconstruction of gene regulatory networks dissects transcriptional control of intratumoral regulatory T cells), and mapping the differentiation vectors of tumor-infiltrating dendritic cells (Qingyu Huang et al. in Deciphering tumor-infiltrating dendritic cells in the single-cell era).
- Tissue Regeneration and Aging: Identifying basal stem cell transition states in human skin (Shuxiong Wang et al. in Single cell transcriptomics of human epidermis identifies basal stem cell transition states), tracking T cell diversion in aging tissues (Meaghan M. Broman in Immune cell single-cell RNA sequencing analyses link an age-associated T cell subset to symptomatic benign prostatic hyperplasia), and characterizing CD4+ T cell memory signatures in melanoma survivors (Jichang Han et al. in 645 Single-cell transcriptional and clonal characterization of CD4+ T cells across tissues in long-term melanoma survivors).
- Organ Failure and Neurodegeneration: Mapping transitional intermediates in the cardiac vascular niche during heart failure (Fabian Peisker et al. in Mapping the cardiac vascular niche in heart failure) and characterizing the directional slide of astrocytes into an "exhausted" state during the progression of Alzheimer’s disease (Alberto Serrano-Pozo et al. in Astrocyte transcriptomic changes along the spatiotemporal progression of Alzheimer’s disease).
By treating the ratio of unspliced to spliced transcripts as a physical clock, RNA velocity bridges the gap between static transcriptomic snapshots and the fluid, continuous reality of cellular biology.
Key Foundational Work
Transcriptome variations among human embryonic stem cell lines are associated with their differentiation propensity (2018) — contributed 1 concept elements to this direction.
Predicting Cellular Growth from Gene Expression Signatures (2009) — contributed 1 concept elements to this direction.