Benchmarking large language models for extracting biobank-derived insights into health and disease
PLOS Computational Biology · 2026
DOI: 10.1371/journal.pcbi.1014224 ↗Kurz gefasst
Fachartikel in der Zeitschrift PLOS Computational Biology, 2026. Verfasst von Manuel Corpas und 1 weiteren Autor:innen. Beteiligte Einrichtung: Life Sciences, University of Westminster, London, United Kingdom. Hauptthemen laut MEDLINE: Large Language Models, Biological Specimen Banks, Benchmarking, Data Mining. Die Arbeit nennt 33 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: South London and Maudsley NHS Foundation Trust, MND Scotland, Motor Neurone Disease Association.
Abstract (Originalsprache)
Biobank-scale datasets such as the UK Biobank have become foundational resources for advancing biomedical discovery. Yet the complexity and heterogeneity of these resources, spanning genomics, imaging, clinical records, and metadata, pose substantial barriers to access and interpretation. Large Language Models (LLMs) offer a promising avenue for making such datasets more navigable through natural language interfaces. However, the extent to which current general-purpose LLMs can retrieve and synthesize biobank-specific insights has not yet been systematically evaluated. In this study, we present a reproducible, multi-metric evaluation framework to benchmark the capabilities of leading LLMs. We evaluated six leading large language models: Gemini 3 Pro, Claude Opus 4.5, Claude Sonnet 4.5, GPT-5.2, Mistral Large 2, and DeepSeek V3, on four benchmark tasks designed to assess biobank-related knowledge retrieval. We evaluate model performance across six dimensions (semantic accuracy, factual correctness, domain knowledge, reasoning quality, response depth, and biobank specificity) and assessed output consistency using curated UK Biobank references and a robust random baseline. All models outperformed the baseline by 2× to 3× , with strong statistical separation (p < 0.001), confirming meaningful biobank-specific knowledge retrieval. Gemini 3 Pro achieved the highest overall accuracy across tasks such as keyword synthesis, institution recognition, and topic inference, while Claude Sonnet 4.5 demonstrated the most uniform performance across evaluation dimensions. Our benchmark provides a rigorous framework for evaluating LLMs in biomedical settings. Using the UK Biobank as a real-world testbed, we highlight both the capabilities and limitations of current models, measuring their capacity to recall structured biomedical knowledge consistent with authoritative biobank metadata.
Schlagwörter
MeSH-Schlagwörter (MEDLINE)
Autor:innen und Institutionen
- Life Sciences, University of Westminster, London, United Kingdom.
- The Alan Turing Institute, London, United Kingdom.
- Department of Biostatistics and Health Informatics, King's College London, London, United Kingdom.
- Department of Basic and Clinical Neuroscience, King's College London, London, United Kingdom.
- Perron Institute for Neurological and Translational Science, Perth, Western Australia, Australia.
- Biomedical Research Centre, South London and Maudsley NHS Foundation Trust, London, United Kingdom.
ORCID-Profile: Manuel Corpas
Förderung
- South London and Maudsley NHS Foundation Trust
- MND Scotland
- Motor Neurone Disease Association
- National Institute for Health and Care Research
- Spastic Paraplegia Foundation
- Rosetrees Trust
- Darby Rimmer MND Foundation
- Medical Research Council
Zitieren
APA
Corpas, M., & Iacoangeli, A. (2026). Benchmarking large language models for extracting biobank-derived insights into health and disease. PLOS Computational Biology. https://doi.org/10.1371/journal.pcbi.1014224
Vancouver
Corpas M, Iacoangeli A. Benchmarking large language models for extracting biobank-derived insights into health and disease. PLOS Computational Biology. 2026. doi:10.1371/journal.pcbi.1014224
BibTeX
@article{Corpas2026,
author = {Corpas, Manuel and Iacoangeli, Alfredo},
title = {Benchmarking large language models for extracting biobank-derived insights into health and disease},
journal = {PLOS Computational Biology},
year = {2026},
doi = {10.1371/journal.pcbi.1014224}
}
EUVIMED