SpatialMultiOmics

Joint NMF factorization for integrative spatial proteomics and transcriptomics analysis.

Overview

SpatialMultiOmics provides a computational framework for integrating spatial transcriptomics (Visium, MERFISH) and spatial proteomics (CODEX, MIBI) data using joint Non-negative Matrix Factorization (NMF). The method decomposes combined spot/cell-level expression matrices into shared cell-type factors and computes spatial co-localization scores using the Jones-Scornecchi statistic.

Features

• Multi-platform support: Works with Visium, MERFISH, Xenium, Stereo-seq (transcriptomics) and CODEX, MIBI, IMC, CyCIF (proteomics)
• Joint NMF factorization: Simultaneous decomposition of combined expression matrices
• Cell-type annotation: Reference-based annotation using marker genes and proteins
• Spatial analysis: Niche composition and Jones-Scornecchi co-localization scoring
• CLI interface: Easy integration into bioinformatics pipelines

Installation

pip install spatialmultiomics

Or install from source:

git clone https://github.com/junior1p/SpatialMultiOmics.git
cd SpatialMultiOmics
pip install -e .

Usage

Python API

from SpatialMultiOmics import run_pipeline

result = run_pipeline(
    transcript_path="transcriptomics.csv",
    protein_path="proteomics.csv",
    platform_tx="Visium",
    platform_prot="CODEX",
    n_factors=15,
    radius=50.0,
    output_dir="results"
)

print(f"Cell types: {result.cell_types}")
print(f"Co-localization: {result.co_localization}")

CLI

python -m SpatialMultiOmics \
    --tx transcriptomics.csv \
    --prot proteomics.csv \
    --tx-platform Visium \
    --prot-platform CODEX \
    --n-factors 15 \
    --radius 50.0 \
    --output results

Output Files

• shared_labels.csv: Cell-type assignments for each spot/cell
• co_localization.csv: Jones-Scornecchi co-localization scores
• niche_composition.csv: Neighborhood composition per cell
• W_transcript.npy, W_protein.npy, H_shared.npy: NMF loadings

Method

1. Data Loading: Load spatial transcriptomics and proteomics data
2. Alignment: Align modalities at cell/spot resolution
3. NMF Decomposition: Joint factorization of combined matrix
4. Annotation: Assign cell types using marker genes/proteins
5. Spatial Analysis: Compute niche composition and co-localization

Citation

[Coming soon]

License

MIT