A knowledge graph approach to predict and interpret disease-causing gene interactions.

Area of Science:

• Genetics
• Bioinformatics
• Computational Biology

Background:

• Understanding gene interactions is crucial for genetic disease research, especially for oligogenic diseases influenced by multiple gene variants.
• Current machine-learning methods for identifying gene combinations in oligogenic diseases lack interpretability, hindering clinical validation.
• There is a need for predictive approaches that offer transparent explanations for disease-causing gene interactions.

Purpose of the Study:

• To develop a novel, interpretable predictive approach for identifying disease-causing gene interactions.
• To provide accurate predictions of pathogenic gene interactions in oligogenic diseases.
• To generate explanations for predictions, aiding medical experts in validation and comprehension.

Main Methods:

• Constructed BOCK, a knowledge graph integrating clinical data, biomedical networks, and ontologies for oligogenic diseases.
• Developed a predictive framework using heterogeneous paths within the knowledge graph to connect gene pairs.
• Trained an interpretable decision set model to predict pathogenic gene interactions and identify associated patterns.

Main Results:

• The BOCK knowledge graph facilitates exploration of disease-causing genetic interactions.
• The predictive framework accurately identifies pathogenic gene interactions and associated disease patterns.
• The approach provides subgraph explanations for each prediction, detailing the evidence used.

Conclusions:

• The developed method leverages knowledge graphs and heterogeneous paths for interpretable prediction of pathogenic gene interactions.
• This approach enhances understanding of molecular mechanisms in oligogenic diseases.
• It represents a novel application of knowledge graphs for transparent and insightful genetic research predictors.