Yu Zhang Yu Zhang

Research

I work toward a future where therapeutics are not just discovered, but also designed: AI models that propose novel molecules while anticipating biological response and host safety. The goal is a unified workflow—generation, prediction, and experimental feedback—so candidates are potent by design and safer by default.

Closed-loop workflow: Generate, Predict, Test, Learn

Closed-loop discovery: generate → predict → test → learn.

Focus areas

What I build and apply day-to-day.

Generative design

Create novel small molecules optimized for target activity and developability—not just rediscover known scaffolds.

Diffusion Property control Novelty

Safety-aware modeling

Treat toxicity as a first-class objective by embedding safety predictors directly into prioritization and design loops.

Multitask Toxicity Selectivity

AMR-focused discovery

Build adaptive ML frameworks to identify novel mechanisms and candidates against resistant Gram-negative pathogens.

AMR Targets Validation

Multi-modal mechanism of action

Integrate structure, transcriptomics, and high-content imaging to infer mechanism and guide experimental follow-up.

RNA-seq Imaging Systems biology

Principles

Safety-first
Efficacy without host safety doesn’t matter.
Biology-grounded
Models should reflect biological context, not just chemistry.
Validation-driven
Design choices are shaped by what we can test and learn.