Tag
WMLLM is a self-evolving optimization-agent framework that uses predict-then-act world modeling with large language models to improve sample efficiency in black-box optimization, achieving state-of-the-art results in multi-objective molecular optimization.
The paper introduces Elite-Weighted Supervised Fine-tuning (EW-SFT), a method for goal-directed molecular optimization that uses reward to guide elite selection and updates via the model's native loss, applicable across various generative architectures and tasks.
Introduces Q-Steer, a rollout-time action-value guidance for molecular policy optimization that uses a frozen prefix-action value scorer to improve sampling without changing the online oracle budget.
PRECEDE is a precedent-guided co-scientist for side-effect-aware drug redesign that frames drug modification as evidence-grounded reasoning over drug–side-effect associations and biomedical knowledge graphs, coordinated by an LLM orchestrator with human oversight.
This paper studies online adaptation strategies for discrete diffusion models in molecular optimization, identifying complementary components like acquisition, reward shaping, debiasing, replay, and validity control that improve feedback efficiency on small-molecule and protein tasks.
Active-GRPO introduces an adaptive imitation and self-improving reasoning framework that dynamically decides when to imitate references and when to reinforce the model's own discoveries for molecular optimization, achieving statistically significant improvements over previous methods on the TOMG-Bench-MolOpt benchmark.
The Nanotechnology Molecular Optimization (NMO) Benchmark introduces physics-based molecular design tasks replacing drug-discovery-focused metrics, aiming to drive scientific discovery in nanotechnology. The paper shows that advanced methods underperform simpler approaches on NMO, and proposes new baseline methods including a novel representation and domain-agnostic pretraining.
This paper introduces PROBE, a framework that uses LLM agents to iteratively optimize ligands in structure-based drug design by probing pocket-ligand complex responses before editing, achieving state-of-the-art results on CrossDocked2020.
ATOM is a multi-agent framework that formulates molecular optimization as a tree-structured search with specialized agents along paths, enabling exploration of alternative molecular trajectories and improving Pareto coverage in multi-objective benchmarks.
TRACE is a trajectory-aware LLM agent for molecular lead optimization that uses sequential decision-making over molecular optimization tools, achieving improved ADMET properties while preserving molecular similarity.