Scholarlink

<p>The lack of potential treatment candidates is a critical issue in managing rare diseases. A knowledge-based approach, incorporating relations and concepts from multiple types of biological databases, can allow for the discovery of new uses for existing drug candidates inexpensively and expeditiously. We have assembled RD<sup>2</sup>-KG, a knowledge graph incorporating various negative and positive biological interactions collected from over 10 publicly available standard biomedical databases. The entities and relations in RD<sup>2</sup>-KG are aligned to ensure data traceability and support AI studies. Only high-quality interactions are represented in RD<sup>2</sup>-KG making it highly reliable and permitting precise predictions. RD<sup>2</sup>-KG has 105,241 nodes and 4,472,526 edges. To validate RD<sup>2</sup>-KG for drug repurposing, link prediction using CompGCN on a subnetwork of drug-disease-gene entities was carried out for Parkinson’s Disease and Muscular Dystrophy. A testable hypothesis for mechanism of action was generated leveraging RD<sup>2</sup>-KG based on the predicted drug-disease associations using network analysis and a custom path-scoring function. The drugs generated and the mechanism of action were validated using published literature. We have also provided guidelines to replicate RD<sup>2</sup>-KG for continuous updates in the future. </p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Assembling a multiscale biomedical knowledge graph for explainable drug repurposing in rare diseases

  • 期刊论文

  • 发表日期：

    2025年5月29日

查看全文
  • A. Arun Kumar Annadurai,
  • Samarth Bhandary,
  • Swathi Gopal Hegde,
  • Jhinuk Chatterjee

摘要

The lack of potential treatment candidates is a critical issue in managing rare diseases. A knowledge-based approach, incorporating relations and concepts from multiple types of biological databases, can allow for the discovery of new uses for existing drug candidates inexpensively and expeditiously. We have assembled RD2-KG, a knowledge graph incorporating various negative and positive biological interactions collected from over 10 publicly available standard biomedical databases. The entities and relations in RD2-KG are aligned to ensure data traceability and support AI studies. Only high-quality interactions are represented in RD2-KG making it highly reliable and permitting precise predictions. RD2-KG has 105,241 nodes and 4,472,526 edges. To validate RD2-KG for drug repurposing, link prediction using CompGCN on a subnetwork of drug-disease-gene entities was carried out for Parkinson’s Disease and Muscular Dystrophy. A testable hypothesis for mechanism of action was generated leveraging RD2-KG based on the predicted drug-disease associations using network analysis and a custom path-scoring function. The drugs generated and the mechanism of action were validated using published literature. We have also provided guidelines to replicate RD2-KG for continuous updates in the future.