Identifying Disease-Gene Associations With Graph-Regularized Manifold Learning.

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Identifying disease-gene associations is crucial for understanding complex diseases, impacting diagnosis, treatment, and prevention.
• Current computational methods are often disease-specific and struggle to predict genes for diseases lacking known associations.
• Existing approaches do not leverage disease similarities, limiting their predictive power.

Purpose of the Study:

• To develop a novel computational method for predicting disease-gene associations.
• To overcome limitations of disease-specific prediction models and address diseases with no prior gene associations.
• To utilize disease and gene similarities within a unified framework.

Main Methods:

• A manifold learning-based approach was developed, mapping diseases and genes into a lower-dimensional space.
• The model assumes shorter geodesic distances between diseases and their associated genes.
• Predictions are made based on calculated geodesic distances between disease-gene pairs, incorporating known associations, disease similarity, and gene similarity.

Main Results:

• Achieved an Area Under the Receiver Operating Characteristic Curve (AUC) of 0.882 for diseases with multiple known genes.
• Achieved an AUC of 0.854 for diseases with only one known associated gene.
• Demonstrated capability in identifying novel disease-gene associations through de novo studies on Lung Cancer and Bladder Cancer.

Conclusions:

• The proposed manifold learning method effectively predicts disease-gene associations, outperforming existing disease-specific models.
• The approach successfully identifies candidate genes for diseases with limited or no prior known associations.
• This method offers a valuable tool for accelerating the discovery of disease-gene links, with implications for clinical applications.