🪿 GEESE

Genotype-aware End-to-End Spatio-temporal Embedding for Behavioral Phenotyping

Behavioral phenotyping of rodent models from 3D pose recordings that predicts behavior over time, cohort, and genotype, with a chat assistant for querying the results.

   

Yiran Ding¹    Yuen Gao²    Chunqi Qian²    Zijun Cui¹

¹Department of Computer Science and Engineering, Michigan State University    ²Department of Radiology, Michigan State University

---

Usage

Open the live demo, then either upload a .mat recording or pick a built-in example, and click Analyze. The app returns:

• a behavior distribution and a behavior timeline,
• cohort and genotype predictions (with probabilities),
• a downloadable Excel summary,
• a chat box for asking questions about the results.

Running locally

pip install -r requirements.txt
python app.py

Data

Example recordings come from the socialDANNCE dataset (Harvard Dataverse, CC0).

Acknowledgements

This work builds on MOMENT, a time-series foundation model.

Citation

@misc{ding2026geesegenotypeawareendtoendspatiotemporal,
      title={GEESE: Genotype-aware End-to-End Spatio-temporal Embedding for Behavioral Phenotyping}, 
      author={Yiran Ding and Yuen Gao and Chunqi Qian and Zijun Cui},
      year={2026},
      eprint={2605.24370},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2605.24370}, 
}