Explainable deep transfer learning model for disease risk prediction using high-dimensional genomic data.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Precision medicine relies on accurate disease risk prediction models.
• High-dimensional genomic data presents challenges due to noise and complex relationships.
• Deep learning models show promise but struggle with dimensionality and interpretability.

Purpose of the Study:

• To develop a biologically interpretable and computationally efficient deep learning framework for disease risk prediction using genomic data.
• To address the limitations of existing deep learning models in handling high-dimensional genomic data and providing biological insights.

Main Methods:

• Developed a group-wise feature importance score for efficient gene selection, detecting both linear and non-linear effects.
• Designed an explainable transfer-learning based deep neural network (DNN) method.
• Integrated feature selection information into the DNN to capture complex predictive effects.

Main Results:

• The proposed DNN framework demonstrated efficient detection of predictive genomic features.
• The method accurately predicted disease risk, outperforming existing approaches in simulations and real-world data analyses.
• The framework provides biological interpretability by focusing on selected predictive genes.

Conclusions:

• The developed DNN framework offers a biologically interpretable and efficient solution for disease risk prediction from genomic data.
• This approach advances precision medicine by improving the accuracy and understanding of genomic-driven disease risk.
• The method is suitable for genome-wide data analysis and overcomes key limitations of traditional deep learning applications in genomics.