RepurposeAD: Knowledge Graph-Guided LLM Reasoning for Alzheimer’s Drug Repurposing
Pharmaceuticals · 04.11.2026 · 19 (11) · 1749
DOI: 10.3390/ph19111749 ↗Kurz gefasst
Fachartikel in der Zeitschrift Pharmaceuticals, online erschienen am 04.11.2026. Verfasst von Shahidur Rahoman Sohag und 6 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: King Abdulaziz University.
Themen: Demenz
Abstract (Originalsprache)
Background/Objectives: Alzheimer’s disease (AD) drug development continues to suffer from high late-stage attrition and narrow mechanistic diversity, which makes the systematic repurposing of approved compounds an urgent priority. Existing computational repurposing methods operate on fragmented biomedical knowledge, rarely integrate transcriptomic and pharmacological evidence within a single reasoning process, and provide little of the mechanistic transparency required for clinical translation. This paper presents RepurposeAD, a knowledge graph-guided large language model (LLM) reasoning framework that identifies and ranks repurposed candidates for AD.
Methods
RepurposeAD constructs a unified AD-centric knowledge graph from protein interaction networks, drug–target databases, clinical trial registries, and transcriptomic signatures, linking genetic risk loci, amyloid and tau pathway nodes, neuroinflammatory mediators, and established therapeutic mechanisms. A constrained multi-hop path reasoning module scores drug-to-disease routes and attributes each route to one of five pathological axes, after which an LLM-based evidence synthesis module reads the retrieved paths, trial records, and literature to produce calibrated mechanistic justifications and a final iterative re-ranking.
Results
On a curated benchmark of 1847 drug–disease associations spanning amyloid, tau, neuroinflammatory, synaptic, and metabolic targets, RepurposeAD attains a top-10 hit rate of 0.83, an area under the precision–recall curve of 0.79, an area under the receiver operating characteristic curve of 0.93, and a mean reciprocal rank of 0.65, exceeding the strongest of five baselines by 0.07 in hit rate and 0.08 in precision–recall area. Three of the ten highest ranked candidates have active or completed trials in AD-adjacent indications.
Conclusions
RepurposeAD provides a transparent, scalable, and mechanistically grounded pipeline for accelerating repurposing in AD and related dementias.
Förderung
- King Abdulaziz University
Zitieren
APA
Sohag, S. R., Pasha, S. M. M., Ghanem, H., Abulfaraj, A. W., Ameen, M., Binzagr, F., & Jumani, A. K. (2026). RepurposeAD: Knowledge Graph-Guided LLM Reasoning for Alzheimer’s Drug Repurposing. Pharmaceuticals, 19(11), 1749. https://doi.org/10.3390/ph19111749
Vancouver
Sohag SR, Pasha SMM, Ghanem H, Abulfaraj AW, Ameen M, Binzagr F, et al. RepurposeAD: Knowledge Graph-Guided LLM Reasoning for Alzheimer’s Drug Repurposing. Pharmaceuticals. 2026;19(11):1749. doi:10.3390/ph19111749
BibTeX
@article{Sohag2026,
author = {Sohag, Shahidur Rahoman and Pasha, Syed Murtoza Mushrul and Ghanem, Hossny and Abulfaraj, Anas W. and Ameen, Mohammed and Binzagr, Faisal and Jumani, Awais Khan},
title = {RepurposeAD: Knowledge Graph-Guided LLM Reasoning for Alzheimer’s Drug Repurposing},
journal = {Pharmaceuticals},
year = {2026},
volume = {19},
number = {11},
pages = {1749},
doi = {10.3390/ph19111749}
}
EUVIMED