PerturbODE (ICML 2026)

PerturbODE is an interpretable Neural ODE framework for joint genetic perturbation prediction and gene regulatory network discovery from single-cell perturbational data. You can refer to our preprint or ICML 2026 poster for more details.

Code with the TF Atlas experiment reproduction will be released soon.

Contact

• Zaikang Lin: zl3135@columbia.edu or zaikang.lin@helmholtz-munich.de
• Sei Chang: shc2170@columbia.edu
• Aaron Zweig: az2888@columbia.edu
• David A. Knowles: dak2173@columbia.edu

Citation

If you use PerturbODE in your research, please cite:

@article{perturbode2026,
  title   = {Interpretable Neural ODEs for Gene Regulatory Network Discovery under Perturbations},
  author  = {Lin, Zaikang and Chang, Sei and Zweig, Aaron and Kang, Minseo and Theis, Fabian J. and Azizi, Elham and Knowles, David A.},
  journal = {arXiv preprint arXiv:2501.02409},
  year    = {2025}
}