CARDBiomedBench

Paper | Github

Dataset Summary

CARDBiomedBench is a biomedical question-answering benchmark designed to evaluate Large Language Models (LLMs) on complex biomedical tasks. It consists of a curated set of question-answer pairs covering various biomedical domains and reasoning types, challenging models to demonstrate deep understanding and reasoning capabilities in the biomedical field.

Data Fields

• question: string - The biomedical question posed in the dataset.
• answer: string - The corresponding answer to the question.
• bio_category: string - The biological category or categories associated with the question. Multiple categories are separated by a semicolon (;).
• reasoning_category: string - The reasoning category or categories associated with the question. Multiple categories are separated by a semicolon (;).
• uuid: string - Unique identifier for the question-answer pair.
• template_uuid: string - Unique identifier of the original template question from which this instance was derived.

Citation

@article {Bianchi2025.01.15.633272,
  author = {Bianchi, Owen and Willey, Maya and Avarado, Chelsea X and Danek, Benjamin and Khani, Marzieh and Kuznetsov, Nicole and Dadu, Anant and Shah, Syed and Koretsky, Mathew J and Makarious, Mary B and Weller, Cory and Levine, Kristin S and Kim, Sungwon and Jarreau, Paige and Vitale, Dan and Marsan, Elise and Iwaki, Hirotaka and Leonard, Hampton and Bandres-Ciga, Sara and Singleton, Andrew B and Nalls, Mike A. and Mokhtari, Shekoufeh and Khashabi, Daniel and Faghri, Faraz},
  title = {CARDBiomedBench: A Benchmark for Evaluating Large Language Model Performance in Biomedical Research},
  year = {2025},
  publisher = {Cold Spring Harbor Laboratory},
  URL = {https://www.biorxiv.org/content/early/2025/01/19/2025.01.15.633272},
  journal = {bioRxiv}
}