Skip to content

Cell Fate Prediction with Cafe

What is Cell Fate Prediction?

Cell fate prediction is the task of inferring the developmental trajectories of cells from single-cell RNA sequencing (scRNA-seq) data. In a developing tissue or differentiating system, cells transition through continuous states — from progenitor cells to fully differentiated cell types. Understanding these transitions is fundamental to developmental biology, disease progression, and regenerative medicine.

The Core Problem

Given a snapshot of gene expression profiles from thousands of individual cells (scRNA-seq data), the goal of cell fate prediction is to:

  1. Recover the underlying developmental topology — the branching structure of how cell states relate to each other.
  2. Order cells along trajectories — assign each cell a pseudotime value representing its position along a differentiation path.
  3. Identify lineage branch points — detect where a progenitor population diverges into distinct terminal fates.
  4. Quantify transition dynamics — measure the direction and speed of cellular transitions (e.g., via RNA velocity).

Why is it Challenging?

  • Static snapshots: scRNA-seq is destructive — we only observe each cell once, never see transitions directly.
  • High dimensionality: Each cell is measured across ~20,000 genes, with substantial technical noise.
  • Complex topologies: Trajectories can be linear, branching, convergent, or cyclic.
  • Method diversity: Dozens of trajectory inference methods exist, each with different assumptions, strengths, and weaknesses.

How Cafe Addresses This Problem

Cafe (Cellular Fate Explorer) provides an integrated platform that unifies the three essential stages of cell fate analysis:

1. Unified Data Structure — FateAnnData

FateAnnData extends AnnData with cafe metadata (uns["cafe"]), auto-detected prior information, and multi-trajectory storage. All methods output into a unified MilestoneNetwork (milestones + progressions + divergence regions). See Data Structure.

2. Trajectory Inference

Diverse methods — pseudotime, end-state probabilities, RNA velocity, optimal transport — run via four backends (Python function, Conda, CFE Docker, Dynverse Docker) and output through one of 11 wrapper types into a unified MilestoneNetwork. See Method Wrappers.

3. Visualization

Embedding-based trajectory curves, method-specific wrapper views, velocity streams, and benchmark comparison plots. See Plot Module.

4. Benchmarking

Topology similarity, cluster assignment accuracy, pseudotime correlation, velocity coherence, and feature importance. See Metric Module.

flowchart TD
    A["<b>① Load Data</b>"] --> B["<b>② Run Method</b>"]
    B --> C["<b>③ Visualization</b>"]
    C --> D["<b>④ Benchmark</b>"]
    C --> E["<b>⑤ Downstream Analysis</b>"]

Task Flow

References