Matching biomedical ontologies with GCN-based feature propagation.

Area of Science:

• Biomedical informatics
• Artificial intelligence
• Computational biology

Background:

• Biomedical ontology matching is crucial for interoperability but challenged by terminology heterogeneity and semantic ambiguity.
• Traditional rule-based methods struggle with complex ontology structures and manual rule design.
• Existing knowledge graph embedding methods focus on entity alignment, not abstract concepts in ontologies.

Purpose of the Study:

• To propose a novel deep learning approach, BioOntGCN, for effective biomedical ontology matching.
• To address the limitations of traditional methods and entity alignment techniques in handling abstract concepts.

Main Methods:

• Generated a pair-wise connectivity graph (PCG) from ontology pairs, with nodes as concept-pairs and edges as property-pairs.
• Employed a convolutional neural network (CNN) to extract node similarity features.
• Utilized a graph convolutional network (GCN) to propagate features and learn concept-pair embeddings for binary classification.

Main Results:

• BioOntGCN achieved state-of-the-art performance on real-world biomedical ontologies from the Ontology Alignment Evaluation Initiative (OAEI).
• The approach significantly outperformed existing entity alignment methods.
• BioOntGCN demonstrated superior effectiveness compared to traditional ontology matching systems.

Conclusions:

• BioOntGCN offers a more effective representation learning-based approach for biomedical ontology matching.
• The method is highly applicable to ontologies with abstract concepts, overcoming limitations of entity alignment.
• This work advances biomedical data interoperability through improved ontology matching.