Disease gene prediction for molecularly uncharacterized diseases

Area of Science:

• Genetics and Bioinformatics
• Computational Biology
• Systems Medicine

Background:

• Network medicine advances disease gene discovery for molecularly characterized conditions.
• Predicting genes for diseases lacking a known molecular basis remains a significant challenge.

Purpose of the Study:

• To introduce Cardigan (ChARting DIsease Gene AssociatioNs), a novel network approach for prioritizing gene-disease associations.
• To enable gene prediction for diseases without a known molecular basis.
• To improve the accuracy of disease gene and module prediction.

Main Methods:

• Utilized semi-supervised learning and disease phenotype similarity measures.
• Employed network medicine principles, including interactome analysis.
• Evaluated performance on OMIM datasets using weighted and binary interactomes, with time-based data splits.

Main Results:

• Cardigan accurately predicts disease genes for molecularly uncharacterized diseases.
• Outperformed state-of-the-art methods by 14%-65% in predicting genes for molecularly characterized diseases.
• Achieved superior performance (87%-299%) in disease module prediction compared to existing approaches.

Conclusions:

• Cardigan offers a robust method for advancing gene-disease association discovery, particularly for challenging uncharacterized diseases.
• The approach significantly improves upon current state-of-the-art methods in both gene and disease module prediction.
• Cardigan holds promise for accelerating the identification of disease genes and understanding disease mechanisms.