Evolving knowledge graph similarity for supervised learning in complex biomedical domains.

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• Biomedical ontologies are crucial for knowledge graphs, but existing data mining methods using vector representations lack full semantic capture.
• Machine learning approaches exploring semantic similarity are promising but require task-specific fine-tuning of ontology perspectives.
• Optimizing semantic similarity aspects for learning tasks is complex and often relies on expert knowledge.

Purpose of the Study:

• To develop an automated method for selecting optimal semantic similarity aspects for knowledge graph-based machine learning tasks.
• To improve the accuracy and applicability of semantic similarity computations in biomedical data mining.

Main Methods:

• Developed evoKGsim, a novel approach employing Genetic Programming to optimize semantic similarity features.
• Utilized Genetic Programming to combine semantic aspects from data for supervised learning tasks.
• Applied the approach to protein-protein interaction prediction using the Gene Ontology knowledge graph.

Main Results:

• evoKGsim outperformed competing strategies, including manually selected combinations of semantic aspects.
• The approach successfully learned species-agnostic models for protein-protein interaction prediction.
• Demonstrated improved prediction for species with limited interaction data.

Conclusions:

• evoKGsim effectively addresses the challenge of expert-driven selection of semantic aspects in knowledge graph applications.
• The methodology shows significant success in protein-protein interaction prediction.
• Paves the way for broader applications of automated semantic similarity optimization.