MolEditRL: Structure-Preserving Molecular Editing via Discrete Diffusion and Reinforcement Learning

Yuanxin Zhuang, Dazhong Shen, Ying Sun · 2025

Read MolEditRL: Structure-Preserving Molecular Editing via Discrete Diffusion and Reinforcement Learning by Yuanxin Zhuang, Dazhong Shen, Ying Sun (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-20131.txt.

Description

Yuanxin Zhuang, Dazhong Shen, and Ying Sun (2025) build a molecular editor that changes a molecule's properties without wrecking its structure. Editing is hard because molecules are graphs, while many editors work on strings or smooth spaces and lose the shape. Their framework, MolEditRL, works in two stages. First, a discrete graph diffusion model learns to rebuild a target molecule from a source one, following instructions. Then reinforcement learning fine-tunes the editor, rewarding property gains and structural fidelity together.

To train and test it, the team built the largest editing dataset yet: three million examples across ten chemical attributes, for single and combined property tasks. MolEditRL beat leading methods on both accuracy and fidelity. Its success rate rose by 74 percent while using 98 percent fewer parameters. The authors credit the pairing of diffusion with reinforcement learning, under graph constraints, for controllability that keeps up with editing power.

Catalog record

CREATORYuanxin Zhuang, Dazhong Shen, Ying Sun
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE70 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2505-20131

Browse the full library · About the Tower of Babel