ReGeNNe: genetic pathway-based deep neural network using canonical correlation regularizer for disease prediction.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Human diseases arise from complex gene-environment interactions.
• Genetic pathway analysis offers deeper biological insights than traditional gene-based methods.
• Understanding these pathways is crucial for accurate disease risk prediction.

Purpose of the Study:

• To propose a novel framework integrating genetic data into pathway structures for enhanced disease risk prediction.
• To leverage an ensemble of convolutional neural networks (CNNs) with Canonical Correlation Regularization.
• To improve interpretability and identify key pathways and genes involved in disease prediction.

Main Methods:

• A two-step framework combining CNNs for intra-pathway gene association extraction.
• Utilizing Canonical Correlation Regularization to fuse features and model inter-pathway interactions.
• Applying the methodology to real-world cancer genetic datasets for validation.

Main Results:

• The proposed deep learning model demonstrated superior performance across multiple cancer prediction tasks.
• Achieved significant Area Under the Curve (AUC) improvements: 11% for kidney cancer staging, 10% for cancer type classification, and 7% for ovarian cancer survival prediction.
• Validated the model's generalizability and robustness on diverse genetic datasets.

Conclusions:

• Deep learning approaches integrating multiple functionally related genes across pathways enhance disease prediction accuracy.
• The framework provides a robust method for disease risk prediction and aids in understanding disease molecular mechanisms.
• The study highlights the potential of advanced computational methods in precision medicine.