Pathway information on methylation analysis using deep neural network (PROMINENT): An interpretable deep learning...

Area of Science:

• Epigenetics
• Computational Biology
• Bioinformatics

Background:

• DNA methylation is a crucial epigenetic regulator of gene expression and phenotype.
• Traditional methods struggle with the nonlinear complexity of DNA methylation data.
• Existing deep learning models lack interpretability and efficiency.

Purpose of the Study:

• To develop an interpretable deep learning method for phenotype prediction using DNA methylation data.
• To integrate biological pathway information for enhanced accuracy and interpretability.
• To improve upon existing methods in terms of prediction accuracy and computational efficiency.

Main Methods:

• Introduced Pathway Information on Methylation Analysis using a Deep Neural Network (PROMINENT).
• Integrated gene-level DNA methylation data with pathway information (GO, KEGG).
• Employed SHapley Additive exPlanations (SHAP) for prioritizing significant genes and pathways.

Main Results:

• PROMINENT demonstrated superior prediction accuracy and computational efficiency across childhood asthma, IPF, and FEP datasets.
• The method successfully identified key genes and pathways implicated in disease mechanisms.
• PROMINENT outperformed existing linear regression and deep learning methods.

Conclusions:

• PROMINENT offers a significant advancement in DNA methylation-based phenotype prediction.
• The method provides high accuracy and interpretability within practical computational constraints.
• PROMINENT holds potential for advancing the understanding of complex diseases and biomedical research.