Related Experiment Video
Updated: Jun 29, 2025
07:35
A Knowledge Graph Approach to Elucidate the Role of Organellar Pathways in Disease via Biomedical Reports
Published on: October 13, 2023
1.6K
MKG-GC: A multi-task learning-based knowledge graph construction framework with personalized application to gastric
Yang Yang1,2, Yuwei Lu2, Zixuan Zheng2
1Computing Science and Artificial Intelligence College, Suzhou City University, Suzhou 215004, China.
Computational and Structural Biotechnology Journal
|April 8, 2024
Summary
A new multi-task learning framework (MKG) extracts gastric cancer knowledge from text, identifying potential drug candidates. Nine of the top ten predicted drugs are already known to treat gastric cancer.
Area of Science:
- Biomedical Informatics
- Computational Biology
- Precision Medicine
Background:
- Vast amounts of biomedical text data are available for precision disease medicine.
- Extracting and utilizing this information for knowledge discovery, particularly for complex diseases like gastric cancer, remains a challenge.
Purpose of the Study:
- To develop and validate a multi-task learning framework (MKG) for automated biomedical knowledge graph construction.
- To apply MKG for extracting gastric cancer (GC)-related knowledge and identifying potential GC drug candidates from literature.
Main Methods:
- Proposed a hard parameter sharing multi-task learning framework (MKG) with three modules: MT-BGIPN (entity recognition), MT-SGTF (entity normalization), and MT-ScBERT (relation classification).
- Utilized bidirectional gated recurrent unit and interactive pointer network for entity recognition, TF-IDF and gated attention for entity normalization, and integrated cross-text, entity, and context features for relation classification.
- Developed a specific gastric cancer knowledge graph (MKG-GC) and employed BioKGE-BERT and a CNN-BiLSTM model for drug-disease prediction.
Main Results:
- Achieved high performance in individual tasks: 84.5% F1 for entity recognition, 94.5% Hits@1 for entity normalization, and 86.9% F1 for relation classification.
- Constructed MKG-GC with 9129 entities and 88,482 triplets.
- Identified potential GC drugs, with nine of the top ten predictions validated against existing literature.
Conclusions:
- The MKG framework effectively automates biomedical knowledge extraction and facilitates the identification of novel therapeutic strategies for gastric cancer.
- The constructed MKG-GC provides a valuable resource for exploring GC-related knowledge and accelerating drug discovery.
- An online platform is available for interactive exploration of the MKG-GC knowledge graph.
