PDGNet: Predicting Disease Genes Using a Deep Neural Network With Multi-View Features.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Understanding phenotype-genotype associations is vital for deciphering disease mechanisms.
• Predicting disease genes using computational methods faces challenges due to sparse and complex medical data.
• Developing efficient deep neural network models for disease gene identification remains a significant hurdle.

Purpose of the Study:

• To develop a novel deep neural network model, PDGNet, for accurate disease gene identification.
• To fuse multi-view features of phenotypes and genotypes for improved prediction.
• To enhance the understanding of disease mechanisms through precise gene identification.

Main Methods:

• Developed PDGNet, a deep neural network model integrating multi-view features of diseases and genes.
• Utilized feedback information from training samples to optimize model parameters.
• Generated deep vector representations for diseases and genes.

Main Results:

• PDGNet demonstrated superior performance compared to state-of-the-art methods, with precision and recall improvements of 9.55% and 9.63%, respectively.
• Predicted genes exhibited strong functional homogeneity and dense interactions with known genes.
• Validated top predicted genes for Parkinson's disease using external data and literature, showing potential for guiding experimental research.

Conclusions:

• PDGNet offers a powerful and efficient approach for disease gene identification.
• The model's ability to integrate diverse data features enhances predictive accuracy.
• The identified candidate genes hold significant potential for future experimental validation and understanding of disease causality.