Related Experiment Video
Updated: Dec 20, 2025
05:47
Evidence-based Knowledge Synthesis and Hypothesis Validation: Navigating Biomedical Knowledge Bases via Explainable AI and Agentic Systems
Published on: June 13, 2025
1.1K
A Method to Learn Embedding of a Probabilistic Medical Knowledge Graph: Algorithm Development
Linfeng Li1,2, Peng Wang3,4, Yao Wang2
1Institute of Information Science, Beijing Jiaotong University, Beijing, China.
JMIR Medical Informatics
|May 22, 2020
Summary
This study introduces PrTransX, a novel method for embedding probabilistic medical knowledge graphs. PrTransX enhances existing TransX algorithms to better represent uncertain relationships, improving accuracy in link prediction tasks.
Area of Science:
- Medical Informatics
- Artificial Intelligence
- Graph Theory
Background:
- Knowledge graph embedding (KGE) represents entities and relations semantically.
- Existing translation-based KGE algorithms (e.g., TransE, TransH) assume deterministic relations, posing challenges for probabilistic medical knowledge graphs.
Purpose of the Study:
- To enhance existing TransX algorithms for probabilistic medical knowledge graphs.
- To incorporate probability values into representation vectors by mapping scores to probabilities and introducing probability-based loss functions.
Main Methods:
- Developed the PrTransX algorithm, an enhancement of TransX models.
- Applied PrTransX to a medical knowledge graph constructed from electronic medical records.
- Evaluated embedding performance using a link prediction task.
Main Results:
- PrTransX outperformed corresponding TransX algorithms across all evaluation metrics.
- Achieved higher top-10 entity prediction accuracy and improved normalized discounted cumulative gain.
- Demonstrated lower mean rank for predicted entities.
Conclusions:
- PrTransX effectively integrates the inherent uncertainty of medical knowledge triplets into embedding vectors.
- The proposed method offers improved semantic representation for probabilistic medical knowledge graphs.
