HGMetaADR: a multimodal heterogeneous graph meta-learning framework for generalized adverse drug reaction prediction
Briefings in Bioinformatics · 02.11.2026 · 27 (6)
DOI: 10.1093/bib/bbag539 ↗Kurz gefasst
Fachartikel in der Zeitschrift Briefings in Bioinformatics, online erschienen am 13.11.2026. Verfasst von Yanran Fu und 7 weiteren Autor:innen. Beteiligte Einrichtung: School of Film, Xiamen University, 422 Siming South Road, 361005 Xiamen, China. Hauptthemen laut MEDLINE: Drug-Related Side Effects and Adverse Reactions, Machine Learning. Die Arbeit nennt 56 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0). Gefördert von: Joint Funds for the Innovation of Science and Technology of Fujian Province, Fujian Provincial Key Laboratory of Environment Factors and Cancer at Fujian Medical University, Public Technology Service Platform Project of Xiamen City.
Abstract (Originalsprache)
Early identification of adverse drug reactions (ADRs), which refer to harmful or unintended responses to medications taken at normal doses, is crucial for reducing clinical trial failures and associated economic costs. However, existing ADR datasets often exhibit a long-tailed label distribution and lack sufficient training examples for rare side effects, making it difficult for vanilla supervised learning models to generalize to underrepresented classes. To address these issues, we propose HGMetaADR, a novel heterogeneous graph-based meta-learning framework for ADR prediction. Built upon a multimodal biomedical knowledge graph, this approach integrates complementary drug-related data, including molecular structures, protein sequences, side effects, and therapeutic indications. It employs a hierarchical attention mechanism to capture semantic dependencies across heterogeneous entities and adopts a model-agnostic meta-learning strategy to enhance adaptation to small-sample classes. We evaluate HGMetaADR under three scenarios—few-shot, zero-shot, and majority—to simulate real-world conditions. Extensive experiments demonstrate that HGMetaADR consistently outperforms state-of-the-art baselines. Furthermore, ablation studies and pathway enrichment analyses reveal the underlying biological mechanisms linking protein targets, therapeutic indications, and side effects, thereby validating the effectiveness of incorporating heterogeneous biomedical knowledge into ADR modeling.
Schlagwörter
MeSH-Schlagwörter (MEDLINE)
Stichwörter der Autor:innen
Autor:innen und Institutionen
- School of Film, Xiamen University, 422 Siming South Road, 361005 Xiamen, China.
- National Institute for Data Science in Health and Medicine, Xiamen University, 422 Siming South Road, 361005 Xiamen, China.
- School of Pharmacy, Jiangxi University of Chinese Medicine, 1688 Meiling Avenue, 330000 Nanchang, China.
- Academy of Military Medical Sciences, 27 Taiping Road, Haidian District, 100850 Beijing, China.
- Department of Epidemiology and Health Statistics, Fujian Provincial Key Laboratory of Environment Factors and Cancer, School of Public Health, Fujian Medical University, No. 1 Xuefu North Road, University Town, Fuzhou,…
- Institute of Advanced Technology and Equipment, Xi'an Jiaotong University, No. 28 Xianning West Road, Beilin District, Xi'an, Shaanxi Province, 710049, China.
ORCID-Profile: Yanran Fu, Ruijiang Li, Kunhong Liu, Jiang Lu, Song He, Xiaochen Bo
Förderung
- Joint Funds for the Innovation of Science and Technology of Fujian Province
- Fujian Provincial Key Laboratory of Environment Factors and Cancer at Fujian Medical University
- Public Technology Service Platform Project of Xiamen City
- National Natural Science Foundation of China
- National Key R&D Program of China
Zitieren
APA
Fu, Y., Wang, M., Li, R., Gao, J., Liu, K., Lu, J., He, S., & Bo, X. (2026). HGMetaADR: a multimodal heterogeneous graph meta-learning framework for generalized adverse drug reaction prediction. Briefings in Bioinformatics, 27(6). https://doi.org/10.1093/bib/bbag539
Vancouver
Fu Y, Wang M, Li R, Gao J, Liu K, Lu J, et al. HGMetaADR: a multimodal heterogeneous graph meta-learning framework for generalized adverse drug reaction prediction. Briefings in Bioinformatics. 2026;27(6). doi:10.1093/bib/bbag539
BibTeX
@article{Fu2026,
author = {Fu, Yanran and Wang, Meijuan and Li, Ruijiang and Gao, Jie and Liu, Kunhong and Lu, Jiang and He, Song and Bo, Xiaochen},
title = {HGMetaADR: a multimodal heterogeneous graph meta-learning framework for generalized adverse drug reaction prediction},
journal = {Briefings in Bioinformatics},
year = {2026},
volume = {27},
number = {6},
doi = {10.1093/bib/bbag539}
}
EUVIMED