velocyto (2018 steady-state model)

中文導讀

velocyto 是 RNA velocity 的原點(La Manno et al., Nature 2018,Linnarsson & Kharchenko), 也是同名的工具(R + Python)。它用 steady-state model:ds/dt = βu − γs,從 unspliced/spliced 比例估 velocity;γ 用 steady-state 細胞(extreme quantiles)迴歸估。對本 wiki 最關鍵的一點: 2018 原版其實是 physically grounded 的——velocity 是真的 time-derivative,timescale 是 hours,而且用 metabolic / EdU labeling 校準到 2.5–3.8 h。這正是 JianhuaXing 說的「sound foundation」,也是後來 scVelo 改成 snapshot-only latent time 之後「丟掉」的東西。

What it is

The original RNA-velocity method and tool (La Manno et al., Nature 2018; source velocyto-2018). The steady-state estimator: fit γ from cells at steady state (u = γs), then read each cell’s unspliced excess/deficit as its velocity.

Model

ds/dt = β·u − γ·s     (β normalized to 1)
steady state:  u = γ·s     →  γ fit by regression on extreme expression quantiles

γ lumps degradation + splicing + gene structure. α is held constant (steady-state assumption) and du/dt = α − βu is not used in the ds/dt estimate (the point JianhuaXing stresses).

Physical-time scorecard

Axisvelocyto (2018)
Latent timenone inferred; velocity is a local physical rate, extrapolation calibrated in hours
Rate scalesteady-state fit gives ratio γ (β≡1); labeling pins the absolute step
External anchoryes — EdU / metabolic labeling (≈2.5–3.8 h extrapolation)
Constant ratesα constant; β≡1; γ gene-specific; honest about multiple-γ failure
Verdictthe physically-grounded origin — velocity in real time (hours), via labeling

See physical-time-grounding; the anchor here is what scVelo dropped and dynamo later restored.

Validated on

Chromaffin / SCP differentiation (lineage tracing), hippocampus, human forebrain neurogenesis, oligodendrocytes, intestinal epithelium, circadian liver. Downstream: Markov random-walk on the velocity field → terminal/root states.

Relation to other methods

velocyto-2018 · RNA velocity · splicing-kinetics-ode · scVelo · dynamo · GraphVelo · metabolic-labeling · physical-time-grounding · velocity-skepticism · HuYizhou · FlowVelo