Saved in:
Bibliographic Details
Main Authors: Zhang, Mingxu, Li, Yuhan, Li, Lujundong, Shen, Dazhong, Xiong, Hui, Sun, Ying
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2605.10831
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910209486618624
author Zhang, Mingxu
Li, Yuhan
Li, Lujundong
Shen, Dazhong
Xiong, Hui
Sun, Ying
author_facet Zhang, Mingxu
Li, Yuhan
Li, Lujundong
Shen, Dazhong
Xiong, Hui
Sun, Ying
contents Large language models possess strong chemical reasoning capabilities, making them effective molecular editors. However, property-relevant information is implicitly entangled across their dense hidden states, providing no explicit handle for property control: a substantial fraction of edits fail to improve or even degrade target properties. To address these issues, we propose SLIM (Sparse Latent Interpretable Molecular editing), a plug-and-play framework that decomposes the editor's hidden states into sparse, property-aligned features via a Sparse Autoencoder with learnable importance gates. Steering in this sparse feature space precisely activates property-relevant dimensions, improving editing success rate without modifying model parameters. The same sparse basis further supports interpretable analysis of editing behavior. Experiments on the MolEditRL benchmark across four model architectures and eight molecular properties show consistent gains over baselines, with improvements of up to 42.4 points.
format Preprint
id arxiv_https___arxiv_org_abs_2605_10831
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle SLIM: Sparse Latent Steering for Interpretable and Property-Directed LLM-Based Molecular Editing
Zhang, Mingxu
Li, Yuhan
Li, Lujundong
Shen, Dazhong
Xiong, Hui
Sun, Ying
Machine Learning
Artificial Intelligence
Computational Engineering, Finance, and Science
Computation and Language
Large language models possess strong chemical reasoning capabilities, making them effective molecular editors. However, property-relevant information is implicitly entangled across their dense hidden states, providing no explicit handle for property control: a substantial fraction of edits fail to improve or even degrade target properties. To address these issues, we propose SLIM (Sparse Latent Interpretable Molecular editing), a plug-and-play framework that decomposes the editor's hidden states into sparse, property-aligned features via a Sparse Autoencoder with learnable importance gates. Steering in this sparse feature space precisely activates property-relevant dimensions, improving editing success rate without modifying model parameters. The same sparse basis further supports interpretable analysis of editing behavior. Experiments on the MolEditRL benchmark across four model architectures and eight molecular properties show consistent gains over baselines, with improvements of up to 42.4 points.
title SLIM: Sparse Latent Steering for Interpretable and Property-Directed LLM-Based Molecular Editing
topic Machine Learning
Artificial Intelligence
Computational Engineering, Finance, and Science
Computation and Language
url https://arxiv.org/abs/2605.10831