Interpretable Neural ODEs for Gene Regulatory Network Discovery under Perturbations

Zaikang Lin^*, Sei Chang^*, Aaron Zweig^*, Minseo Kang, Fabian J. Theis, Elham Azizi, David A. Knowles

^* Equal contribution

International Conference on Machine Learning (ICML), 2026

We present PerturbODE, a framework combining interpretable neural ordinary differential equations with causal graphical models to model cell state dynamics under genetic perturbations. PerturbODE extracts underlying gene regulatory networks from neural ODE parameters, enabling downstream simulation of unseen genetic interventions by grouping genes into interpretable co-regulated modules.

Towards Identifiability of Interventional Stochastic Differential Equations

Aaron Zweig, Zaikang Lin, Elham Azizi, David Knowles

Conference on Uncertainty in Artificial Intelligence (UAI), 2026

We investigate conditions under which the parameters of stochastic differential equations can be uniquely recovered from observational data under multiple interventions. We establish the first theoretical guarantees for parameter recovery using stationary distribution samples, providing exact bounds for linear SDEs and upper bounds for nonlinear cases with small noise.