MolWorld: Molecule World Models for Actionable Molecular Optimization

2026-05-09

AI Summary

This work addresses the challenge in molecular optimization that candidate molecules often lack structural tractability—meaning they cannot be readily derived from known molecules through rational, local transformations. To overcome this limitation, the authors propose a sequential optimization framework grounded in molecular transformation graphs and a learned world model. This approach uniquely integrates molecular property optimization with structural accessibility by iteratively anchoring contextual information, generating new candidates, and updating a learnable representation of the molecular world. The resulting framework enables interpretable and chemically actionable molecular design. Experimental results demonstrate that the method not only discovers high-performing molecules in property optimization and molecular docking tasks but also significantly enhances their structural connectivity to known compounds.

Abstract

Molecular optimization in drug discovery aims to discover molecules with improved target properties, but practical lead optimization often requires more than high predicted scores. A useful candidate should also be actionable: it should be reachable from known molecules through valid local structural transformations, so that it can be interpreted as a plausible revision within an evolving chemical series. Existing de novo and single-molecule optimization methods do not explicitly model such reachability, especially when both the target molecules and the intermediate molecules connecting them to known compounds are unknown. In this work, we formulate actionable molecular optimization as sequential expansion of a molecule-transfer graph, where nodes are molecules and edges encode valid local transformations. We propose MolWorld, a molecule world model-guided framework that treats the current molecule-transfer graph as an evolving search state. At each iteration, MolWorld selects local anchor contexts, generates candidate molecules conditioned on these contexts, evaluates their properties, and uses a learned world model to update the evolving molecule world by retaining admissible candidates and inserting them into the molecule-transfer graph. The expanded molecule world then guides subsequent optimization. Experiments on property optimization and docking-based tasks show that MolWorld discovers high-property molecules while maintaining substantially stronger structural connectivity, supporting actionable and sequential molecular design.

Problem

• actionable molecular optimization
• molecule-transfer graph
• structural connectivity
• lead optimization
• molecular design

Innovation

• molecule world model
• actionable molecular optimization
• molecule-transfer graph
• sequential molecular design
• local structural transformations