Introduction

scCS is a supervised framework for quantifying cell-fate commitment at a biologically annotated furcation. It is designed for analyses in which the researcher already has a justified root population, two or more candidate fate populations, a continuous ordering, and an RNA-velocity transition graph.

Overview of the scCS workflow

What question does scCS answer?

scCS asks:

Given a supplied root and candidate fates, which futures are supported by the velocity graph, how resolved is each cell’s future, and is the cell moving forward or backward along the supplied biological ordering?

scCS does not discover topology, choose terminal states, replace RNA velocity, or claim that every annotated fate is a monotonic outward trajectory. It makes the supervised assumptions explicit and then quantifies their consequences.

Required inputs

Input

Purpose

Typical source

RNA-velocity transition graph

Defines directed future transitions in the original biological manifold

scVelo velocity graph or another row-stochastic transition matrix

Root annotation

Defines the shared incoming population

Curated cell-type or state annotation

Fate annotations

Defines the candidate supervised outcomes

Curated terminal or branch populations

Continuous ordering

Orders cells, selects late anchors, and defines signed progression

Latent time, velocity pseudotime, diffusion pseudotime, CytoTRACE-derived ordering, or experimental time

Conditions and replicates, optional

Enables replicate-first PairScorer or MultiScorer inference

Genuine biological metadata

Core outputs

The recommended graph-based mode is Discounted Future-Fate Propagation (DFFP). It separates quantities that are often collapsed into one vague “commitment score”:

  • Conditional Fate Affinity (CFA): relative future identity among the supplied fates;

  • Discounted Fate Reach (DFR): probability resolved into any supplied fate;

  • Future-Fate Specificity (FFS): decisiveness of the conditional fate distribution;

  • Resolved Commitment (RC): DFR multiplied by FFS;

  • Unresolved Future Probability (UFP): probability not assigned before geometric stopping;

  • Signed Ordering Flux (SOF): expected forward or retrograde movement along the ordering.

Interpretation of reach, specificity, and signed progression

A cell can have strong CFA for a fate and negative SOF. That means the cell’s future identity remains associated with that fate while its local motion is retrograde. Turning Alpha, loop-like Epsilon, unusual Delta, and retrograde Gut are therefore interpretable outcomes rather than errors that must be forced outward.

Which scorer should I use?

Scientific design

Scorer

Inference level

One dataset or one condition

SingleScorer

Cell and population summaries

Two independent conditions

PairScorer

Biological-replicate permutation and bootstrap

Three or more independent conditions

MultiScorer

Omnibus tests, post-hoc comparisons, and planned contrasts

Which scoring mode should I use?

scoring_mode="future_fate" is recommended when the main question is future identity and resolution. It propagates probability on the original velocity graph and does not manufacture a scientific star-space velocity vector.

scoring_mode="instantaneous" asks where immediate transition-induced motion points in the supervised geometry. It is useful for local direction and visualization diagnostics, but it answers a different question.

Worked interpretation

Suppose a cell has CFA [Alpha=0.10, Beta=0.75, Delta=0.05, Epsilon=0.10], DFR 0.55, FFS 0.68, and SOF -0.03.

  • Beta is the dominant supervised future.

  • Only 55% of discounted probability reaches a supplied fate, so the future is not fully resolved.

  • The fate distribution is fairly specific, but not deterministic.

  • RC is 0.55 × 0.68 = 0.374.

  • Negative SOF indicates retrograde motion along the supplied ordering; it does not erase the Beta future association.

Where to go next

Read Mathematical framework for the equations, Quick start for the minimal API workflow, Method selection for the alternatives evaluated, and the Single-condition analysis, Two-condition analysis with PairScorer, and Multi-condition analysis with MultiScorer guides for complete pancreas and Schwann analyses.