AI Paper Atlas

Research map / Probabilistic and scientific modeling / Causal discovery

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

ICML 2026 · Probabilistic and scientific modeling › Causal discovery · approach: Differentiable and score-based DAG learning · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
learn tunable overexpression strength (PerturbODE*) instead of fixed (ablation)PerturbODE (imperfect interv) · TF Atlas human reference GRNsrecall: 0.3659 -> 0.4976Improved
use imperfect rather than perfect intervention modeling (ablation)PerturbODE · TF Atlas human reference GRNsrecall: 0.3561 -> 0.3659Improved
increase number of modules from 10 to 350 (ablation)PerturbODE · AUPRC, TF Atlas vs ChIP-Atlas (223 genes)AUPRC: 0.017528 -> 0.022517Improved
increase number of modules from 350 to 400 (ablation)PerturbODE · AUPRC, TF Atlas vs ChIP-Atlas (223 genes)AUPRC: 0.022517 -> 0.021096No clear effect
switch ODE solver from dopri5 to euler (ablation)PerturbODE · AUPRC, ChIP-Atlas (223 genes)AUPRC: 0.017 -> 0.018No clear effect
lower learning rate from 1e-2 to 1e-4 (ablation)PerturbODE · AUPRC, TF AtlasAUPRC: 0.021 -> 0.022No clear effect
raise learning rate to 1e-1 or 1 (ablation)PerturbODE · AUPRC, TF AtlasAUPRC: 0.022 -> OOMHurt
use PerturbODE instead of GENIE3 at medium thresholdGENIE3 · ChIP-Atlas reference GRNFDR (%): 54.1 -> 52.5Improved
use PerturbODE instead of GENIE3 at low thresholdGENIE3 · ChIP-Atlas reference GRNFDR (%): 54.1 -> 55.4Hurt
use PerturbODE instead of GEARS for held-out intervention predictionGEARS · TF Atlas held-out interventionsW2 distance: 65 -> 84Hurt
use PerturbODE instead of scGPT for held-out intervention predictionscGPT · TF Atlas held-out interventionsW2 distance: 91 -> 84Improved
use PerturbODE instead of NO-TEARS for held-out intervention predictionNO-TEARS · TF Atlas held-out interventionsW2 distance: 396 -> 84Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy