scDynOmics

scDynOmics is an optimized, pretrainable transformer model designed for representation learning from multimodal single-cell data. Motivated by gene regulatory networks, the framework utilizes a Linformer-style attention mechanism to efficiently scale to coding-genome-wide multimodal inputs.

Install

git clone https://github.com/KlughammerLab/scDynOmics.git
cd scDynOmics
pip install .

Test

Note: At least one GPU is recommended for testing and operation.

You can run the built-in test script to verify the installation and core pipeline. From the repository root, run:

python scripts/test.py --data_dir ./data --log_dir ./logs/test/

Tutorial

The documentation and tutorial notebooks are available at here.

Reference

If you use scDynOmics in your research, please consider citing our preprint:

scDynOmics: An Optimized Transformer Model for Representation Learning from Single-Cell Multiomics
Gang Yu, Timothy J.S. Ramnarine, Johanna Klughammer, Simon W. Mages. bioRxiv 2026.

@misc{yu2026scdynomics,
  title={scDynOmics: An Optimized Transformer Model for Representation Learning from Single-Cell Multiomics},
  author={Yu, Gang and Ramnarine, Timothy J. S. and Klughammer, Johanna and Mages, Simon W.},
  year={2026},
  publisher={bioRxiv},
  url={https://doi.org/10.64898/2026.02.28.708160},
  doi={10.64898/2026.02.28.708160}
}