ATOM's multi-agent tree search boosts molecular optimization across conflicting objectives

Researchers from multiple Chinese universities propose ATOM (Agents on a Tree), a multi-agent framework for multi-objective molecular optimization. Traditional methods rely on a single policy or fixed scalarization, limiting trade-off exploration. ATOM instead models the optimization as a tree where each node is an atomic operation and hosts an agent specialized for a particular objective or decision context. Agents coordinate along different paths rather than enforcing global consensus, allowing the system to maintain and compare alternative molecular evolution trajectories. A global memory of past behaviors further balances exploration and exploitation across objectives. This tree-structured interaction enables reasoning over long-horizon dependencies inherent in molecular design.

ATOM was tested on challenging multi-objective benchmarks involving activity, synthesizability, and ADMET-related properties. It consistently achieves improved Pareto coverage and hypervolume over strong baselines, demonstrating the effectiveness of pathwise multi-agent coordination. The code is publicly available on GitHub. By enabling diverse trade-off exploration and long-horizon reasoning, ATOM could accelerate drug discovery and materials design where conflicting objectives must be simultaneously optimized.