A reproducible workflow for analyzing multiplexed imaging data at single-cell resolution using Akoya CODEX/PhenoCycler data.
This project explores the analysis of multiplexed imaging data from tissue samples using cell-level segmentation and protein-expression measurements. The workflow was developed as a technical reproduction of selected analyses from a published spatial proteomics study.
The goal was to start from raw multi-channel images and cell-segmentation masks, extract single-cell features, perform quality control and cell-population identification, and reproduce downstream analyses of cell-type composition and spatial relationships between cell populations.
The project focuses on the computational workflow rather than reproducing the original biological study in its entirety.
The analysis follows a general spatial proteomics workflow:
**Multiplexed images + segmentation masks**
↓
**Single-cell feature extraction**
↓
**Quality control + filtering**
↓
**Dimensionality reduction**
↓
**Cell clustering**
↓
**Cell-type phenotyping**
↙ ↘
**Cell-type composition** **Spatial relationship analysis**
Multi-channel CODEX/PhenoCycler images are combined with corresponding cell-segmentation masks to generate a cell-by-feature representation of the tissue.
For each segmented cell, protein-expression features are extracted from the corresponding image regions.
Cell-level measurements are filtered to remove potential technical artefacts, including:
- Very small segmented objects
- Abnormally large objects
- Fluorescence artefacts
- Other measurements inconsistent with expected cell-level profiles
This step is important because segmentation and imaging artefacts can propagate into downstream clustering and spatial analyses.
The resulting single-cell feature matrix is analyzed using dimensionality reduction and unsupervised clustering.
The workflow includes:
- Dimensionality reduction
- UMAP visualization
- Unsupervised clustering
- Cluster characterization and phenotyping
The objective is to identify groups of cells with similar multiplexed protein-expression profiles and assign biological identities based on their marker profiles.
Identified cell populations are summarized across tissue samples to examine differences in cellular composition.
This provides a higher-level view of the cellular organization of the analyzed tissue.
Cell coordinates are retained throughout the analysis to enable investigation of spatial relationships between cell populations.
Cell-type distances are calculated and summarized to characterize the spatial organization of different cellular populations within the tissue.
The workflow was developed to reproduce selected analyses from Figure 4 of the reference study, including:
- Figure 4C: dimensionality reduction, clustering, and cell-population phenotyping
- Figures 4E and 4H: cell-type proportions
- Figures 4F and 4I: spatial relationships between cell types
The exact implementation may differ from the original study where appropriate, as the objective was to reproduce the analytical results using an independently developed workflow.
The original analysis exercise provided two tissue microarray (TMA) cores containing multiplexed imaging data and corresponding segmentation masks.
The project does not redistribute the original dataset. Users wishing to reproduce the analysis should obtain the appropriate data from the original source or study.
spatial-proteomics-codex/
│
├── data/ # Local analysis data (not distributed)
├── notebooks/ # Exploratory analysis
├── src/ # Reusable analysis code
├── README.md
└── ...
The exact repository structure may evolve as the workflow is cleaned up and generalized.
The analysis uses the Python scientific computing ecosystem for image-based and single-cell analysis.
Key concepts and tools include:
- Multiplexed imaging
- Single-cell feature extraction
- Image segmentation
- Quality control
- Dimensionality reduction
- Unsupervised clustering
- Cell phenotyping
- Spatial analysis
- UMAP visualization
Multiplexed imaging technologies such as CODEX/PhenoCycler provide measurements of many proteins across thousands of individual cells while preserving their spatial organization.
This creates a computational problem that combines image analysis, single-cell analysis, and spatial biology.
This project provided practical experience working across these layers:
Image data → cells → molecular profiles → cell populations → spatial organization
The resulting workflow demonstrates how raw multiplexed imaging data can be transformed into interpretable single-cell and spatial representations.
The analysis was based on the following published study:
Tunable PhenoCycler imaging of the murine pre-clinical tumour microenvironments Abraham et al. (2024)
This repository represents an independent reproduction and computational analysis exercise. It is not intended to reproduce the complete analysis pipeline or biological conclusions of the original study.