End-to-end interpretable disease-gene association prediction

Area of Science:

• Bioinformatics
• Computational Biology
• Genetics

Background:

• Identifying disease-gene associations is crucial for understanding disease mechanisms and developing treatments.
• Experimental verification of these links is costly and time-consuming.
• Existing computational methods often rely on single data sources or require manual meta-path definition, limiting their scope and accuracy.

Purpose of the Study:

• To develop an advanced deep learning model for accurate and interpretable disease-gene association prediction.
• To overcome limitations of existing methods by integrating multi-source heterogeneous data.
• To automatically capture latent interactions between diseases and genes without manual meta-path definition.

Main Methods:

• Proposed a novel end-to-end model named Disease-Gene association Prediction with Parallel Graph Transformer Network (DGP-PGTN).
• Deeply integrated heterogeneous information from diseases, genes, ontologies, and phenotypes.
• Utilized a parallel graph transformer network architecture to automatically learn network representations.

Main Results:

• DGP-PGTN significantly outperformed state-of-the-art methods in disease-gene association prediction.
• The model demonstrated the ability to automatically capture implicit relationships without manual meta-path definition.
• Achieved high accuracy and interpretability in predicting causal relationships.

Conclusions:

• DGP-PGTN offers a powerful and efficient approach for disease-gene association prediction.
• The model's ability to integrate diverse data and learn automatically enhances its applicability in biomedical research.
• This method holds promise for accelerating the discovery of genetic disease mechanisms and therapeutic targets.