TRACE: multi-step lead optimization for drug discovery
Researchers propose TRACE, an agentic framework that treats lead optimization as a sequential decision problem rather than single-step molecular refinement.
The approach balances ADMET (absorption, distribution, metabolism, excretion, toxicity) property improvement with preservation of binding affinity to disease targets—a multi-objective trade-off that one-shot optimization methods miss.
Agentic tool planning lets the system reason about long-term consequences of each structural change, addressing a key bottleneck in accelerating drug discovery pipelines.