Relation Prediction of Co-Morbid Diseases Using Knowledge Graph Completion.

Area of Science:

• Bioinformatics
• Computational Biology
• Medical Informatics

Background:

• Co-morbid diseases increase mortality risk compared to single diseases.
• Accurate prediction of co-morbid disease pairs is essential for improved patient care.
• Existing methods for co-morbid disease prediction lack focus on knowledge graph embedding with tensor factorization.

Purpose of the Study:

• To propose a novel tensor factorization approach for predicting co-morbid disease pairs.
• To introduce complex-valued embeddings within biological knowledge graphs for disease association prediction.
• To leverage biological and biomedical entities for enhanced co-morbidity prediction.

Main Methods:

• Construction of a biological knowledge graph integrating disease-gene associations and background information.
• Application of complex-valued embedding tensor decomposition (ComplEx) for predicting disease associations.
• Utilizing the Markov Clustering (MCL) algorithm on a disease-gene-gene network to identify novel disease pairs.

Main Results:

• Demonstrated the efficacy of complex-valued tensor factorization for co-morbid disease pair prediction.
• Successfully identified new prevalent disease pairs through network analysis.
• Validated the inter-relations of identified disease pairs using an edge prediction task.

Conclusions:

• The proposed complex-valued embedding tensor factorization method offers a promising advancement in predicting co-morbid diseases.
• This approach enhances the understanding of complex disease relationships within biological knowledge graphs.
• The findings can contribute to more accurate risk stratification and personalized medicine strategies.