SupraBench: Molecular Identification (VQA)

Auxiliary vision task of SupraBench: given the 2D depiction of a supramolecular host or guest, identify the molecule (its name / alias set, and where available its canonical SMILES). It probes whether multimodal LLMs can ground chemical structure from images alone.

Links

• 📄 Paper: arXiv:2606.13477
• 💻 Code: https://github.com/Tianyi-Billy-Ma/SupraBench
• 🤗 All datasets: https://huggingface.co/SupraBench

Contents

identification.csv indexes 1,777 molecule images stored under images/.

column	description
molecule_id	integer id; image lives at images/<molecule_id>.png
image	relative path to the 2D depiction (images/<id>.png)
role	host or guest
names_alias_set	accepted name / alias set for the molecule
cano_smiles	canonical SMILES (empty where unresolved)

Usage

from huggingface_hub import snapshot_download
import pandas as pd, os

root = snapshot_download("SupraBench/vqa", repo_type="dataset")
df = pd.read_csv(os.path.join(root, "identification.csv"))
row = df.iloc[0]
print(row["role"], row["names_alias_set"], row["cano_smiles"])
# image path: os.path.join(root, row["image"])

Sources & license

Molecular structures use PubChem and OPSIN; binding records derive from SupraBank (CC-BY-4.0). Released under CC-BY-4.0.

Citation

@article{ma2026suprabench,
  title   = {SupraBench: A Benchmark for Supramolecular Host--Guest Chemistry Reasoning in Large Language Models},
  author  = {Ma, Tianyi and Ma, Yijun and Wang, Zehong and Sun, Weixiang and Li, Ziming and Schmidt, Connor R. and Zhang, Chuxu and Webber, Matthew J. and Ye, Yanfang},
  year    = {2026},
  eprint        = {2606.13477},
  archivePrefix = {arXiv},
  journal = {arXiv preprint arXiv:2606.13477}
}