![[2508.01055] FGBench: A Dataset and Benchmark for Molecular Property Reasoning at Functional Group-Level in Large Language Models](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2508.01055] FGBench: A Dataset and Benchmark for Molecular Property Reasoning at Functional Group-Level in Large Language Models
Summary
FGBench introduces a dataset for molecular property reasoning at the functional group level, enhancing the capabilities of large language models (LLMs) in chemistry tasks.
Why It Matters
This research addresses a gap in existing datasets by focusing on functional group-level information, which can improve the interpretability and performance of LLMs in molecular design and drug discovery. By highlighting the limitations of current models, it sets the stage for future advancements in AI applications within chemistry.
Key Takeaways
- FGBench comprises 625K molecular property reasoning problems with functional group data.
- The dataset includes regression and classification tasks across 245 functional groups.
- Current LLMs show limitations in FG-level property reasoning, indicating a need for improvement.
- The methodology can serve as a framework for future dataset generation in molecular property reasoning.
- FGBench aims to enhance the understanding of structure-property relationships in molecules.
Computer Science > Machine Learning arXiv:2508.01055 (cs) [Submitted on 1 Aug 2025 (v1), last revised 16 Feb 2026 (this version, v4)] Title:FGBench: A Dataset and Benchmark for Molecular Property Reasoning at Functional Group-Level in Large Language Models Authors:Xuan Liu, Siru Ouyang, Xianrui Zhong, Jiawei Han, Huimin Zhao View a PDF of the paper titled FGBench: A Dataset and Benchmark for Molecular Property Reasoning at Functional Group-Level in Large Language Models, by Xuan Liu and 4 other authors View PDF HTML (experimental) Abstract:Large language models (LLMs) have gained significant attention in chemistry. However, most existing datasets center on molecular-level property prediction and overlook the role of fine-grained functional group (FG) information. Incorporating FG-level data can provide valuable prior knowledge that links molecular structures with textual descriptions, which can be used to build more interpretable, structure-aware LLMs for reasoning on molecule-related tasks. Moreover, LLMs can learn from such fine-grained information to uncover hidden relationships between specific functional groups and molecular properties, thereby advancing molecular design and drug discovery. Here, we introduce FGBench, a dataset comprising 625K molecular property reasoning problems with functional group information. Functional groups are precisely annotated and localized within the molecule, which ensures the dataset's interoperability thereby facilitating further mult...
