CLEANMOL: Improving Chemical Understanding of LLMs via SMILES Parsing

Overview

We address limitations in how large language models interpret molecular structures encoded in SMILES format. We introduce CLEANMOL, a framework converting SMILES parsing into structured tasks designed to enhance graph-level molecular comprehension. The approach spans from subgraph to global graph matching with adaptive difficulty scoring. Results demonstrate improved structural understanding and competitive performance on the Mol-Instructions benchmark.

BibTeX

@inproceedings{jang-etal-2025-improving,
  title = "Improving Chemical Understanding of {LLM}s via {SMILES} Parsing",
  author = "Jang, Yunhui  and
    Kim, Jaehyung  and
    Ahn, Sungsoo",
  editor = "Christodoulopoulos, Christos  and
    Chakraborty, Tanmoy  and
    Rose, Carolyn  and
    Peng, Violet",
  booktitle = "Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing",
  month = nov,
  year = "2025",
  address = "Suzhou, China",
  publisher = "Association for Computational Linguistics",
  url = "https://aclanthology.org/2025.emnlp-main.791/",
  doi = "10.18653/v1/2025.emnlp-main.791",
  pages = "15683--15698",
  ISBN = "979-8-89176-332-6"
}