SMODER

SMODER is a Python toolkit for spatial multi-omics deconvolution.

It is designed to integrate spatial transcriptomics with an additional spatial modality, such as protein or chromatin accessibility signals, in order to infer cell-type composition across spatial locations.

The current implementation provides a modular workflow for:

• data preprocessing
• feature engineering
• graph construction
• model training
• deconvolution result generation
• postprocessing reconstruction
• spatial result visualization

The package is organized around reusable components for preprocessing, modeling, configuration, and end-to-end pipeline execution.

Highlights

• Modular Python package structure
• End-to-end runnable spatial multi-omics pipeline
• Support for RNA + second-modality workflows
• Reusable preprocessing, model, and pipeline components
• Tutorial notebook and documentation website support

Project Status

What is currently available:

• the smoder package can be installed and imported
• the current mouse brain H3K27ac pipeline can run end-to-end
• postprocessing utilities are available for denoised RNA and second-modality reconstruction
• visualization utilities are available for spatial heatmaps and embedding-based clustering
• tutorial and API documentation are available through Read the Docs
• output files can be generated successfully, including:
  • cell_type_proportions.csv
  • spatial_decon_result.h5ad
  • training_log.txt

Current recommended execution method:

python -m smoder.pipelines.mousebrain_h3k27ac

Repository Structure

SMODER/
├── smoder/
│   ├── __init__.py
│   ├── config/
│   ├── models/
│   ├── pipelines/
│   ├── preprocessing/
│   ├── postprocessing/
│   └── visualization/
├── data/
├── docs/
├── scripts/
├── tutorials/
├── pyproject.toml
├── requirements.txt
└── README.md

Installation

Requirements

The current validated environment is:

• Linux
• conda-based Python environment
• Python 3.12

The main tutorial and recommended workflow are currently oriented toward this environment.

Install from PyPI

pip install smoder

Install from GitHub

git clone https://github.com/HokazeJunko/SMODER.git
cd SMODER
pip install -e .

Create a Python environment

A dedicated environment is recommended.

For example:

conda create -n smoder python=3.12
conda activate smoder

Data

The current validated example uses a mouse brain RNA + peak dataset.

Required input files:

• sc_mousebrain_processed.h5ad: single-cell reference RNA
• RNA.h5ad: spatial RNA
• peak.h5ad: spatial peak data

Example directory layout:

data/
└── mousebrain_H3K27ac/
   ├── sc_mousebrain_processed.h5ad
   ├── RNA.h5ad
   └── peak.h5ad

How to obtain the data

The validated example dataset is not currently distributed through a public download link.

Please contact the maintainer to obtain the example files.

Additional notes on data preparation are provided in data/README.md and in the documentation site.

Quick Start

Run the current validated pipeline:

python -m smoder.pipelines.mousebrain_h3k27ac

Typical outputs include:

• cell_type_proportions.csv
• spatial_decon_result.h5ad
• training_log.txt

Documentation

Documentation is available through Read the Docs:

https://smoder.readthedocs.io/en/latest/

It currently includes:

• installation guide
• data preparation guide
• tutorial notebook
• result visualizations
• postprocessing and visualization utilities
• API reference

Roadmap

Planned next steps include:

• additional runnable pipelines for other datasets
• improved package interfaces
• more complete dataset instructions
• continued documentation improvements

Citation

If you use SMODER in your work, please cite the corresponding paper or project release once formal citation information becomes available.

Contact

Maintainer: Shucun Xiong