Skip to main content

CellType-Refinery

A comprehensive toolkit for cell-type annotation and refinement from spatial proteomics data.

Overview​

CellType-Refinery is a Python package designed for processing multiplexed imaging data (CODEX, PhenoCycler, Imaging Mass Cytometry) through a multi-stage workflow. It provides hierarchical marker-based annotation, iterative refinement, and comprehensive downstream analysis.

Pipeline Architecture​

Preprocessing Details​

Key Features​

Hierarchical Annotation​

Assign cell types using customizable marker panels with multi-level gating. The system traverses a marker hierarchy, scoring each cell against defined marker sets and applying gating thresholds.

Iterative Refinement​

Improve annotations through automatic and manual curation policies. The refinement engine supports:

  • Subclustering: Re-cluster ambiguous populations at finer resolution
  • Merging: Combine over-split clusters
  • Overrides: Manual corrections for known cell types
  • Relabeling: Update labels without re-clustering

Tissue-Agnostic Design​

Configure CellType-Refinery for any tissue type via YAML templates and JSON marker maps. No code changes required to adapt to new tissues.

Comprehensive Analysis​

  • Composition: Cell-type statistics, diversity metrics (Shannon, Simpson)
  • Spatial: Neighborhood enrichment, Moran's I, cell-type interactions

Package Structure​

celltype_refinery/
├── core/
│   ├── preprocessing/    # Data loading, QC, normalization
│   ├── clustering/       # Leiden clustering
│   ├── annotation/       # Marker-based annotation
│   ├── refinement/       # Iterative refinement
│   ├── consolidation/    # Final label assignment
│   ├── composition/      # Composition statistics
│   ├── spatial/          # Spatial analysis
│   └── review/           # QC flagging
├── pipeline/             # Orchestration
├── io/                   # I/O utilities
├── viz/                  # Visualization
└── cli/                  # Command-line interface