01
PerturbRx
Learns drug- and dose-conditioned state transitions from single-cell perturbation data and transfers them to patient treatment-response prediction.
Yoshitaka Inoue · UMN / NIH
Learning how biological systems respond to therapeutic intervention.
I develop AI models of treatment-conditioned biological state transitions, with the goal of transferring what we learn from molecular and cellular perturbations to patient-level therapeutic response.
Research
Modeling biological dynamics under intervention
My central question is how a biological system changes after an intervention—and which aspects of those dynamics are transferable across drugs, cellular contexts, modalities, and patients.
Biological state
& context
+
& context
Therapeutic
intervention
intervention
State transition
& dynamics
& dynamics
Transfer across contexts
Predict therapeutic response
Reason about mechanism & evidence
Better intervention
models
models
Selected work
From therapeutic response to biological dynamics
02
drGT
Predicts drug response on a drug–cell–gene graph while exposing gene-level contributors to each prediction.
03
DrugAgent
Integrates model predictions, biomedical knowledge, and literature evidence to assess drug–target relationships under conflicting evidence.
Selected publications
Representative papers
- 2026PerturbRx: Treatment-Conditioned Biological State Transitions for Drug Response Prediction
- 2026drGT: Interpretable Drug Response Prediction with Attention-Guided Gene Attribution on a Drug-Cell-Gene Heterogeneous Graph
- 2026DrugAgent: Reliable Multi-Agent Aggregation under Conflicting Biomedical Evidence
- 2026TRPM4 Expression as a Predictive Biomarker and a Mechanistic Driver of Acetalax Activity in Prostate Cancer: Preclinical Efficacy Studies
Recent
PerturbRx preprint released.
ISMB 2026 oral presentations for drGT and DrugAgent.
CCR-FYI 2026 Outstanding Postgraduate Fellow finalist.