Alzheimer's disease prediction using deep learning and XAI based interpretable feature selection from blood gene...

Area of Science:

• Biomedical Informatics
• Neuroscience
• Genomics

Background:

• Alzheimer's disease (AD) is a growing neurodegenerative disorder with a need for accessible early detection.
• Current diagnostic methods for AD are often invasive and expensive.
• Blood gene expression presents a promising, less invasive alternative biomarker source.

Purpose of the Study:

• To develop a comprehensive early detection method for AD using blood gene expression biomarkers.
• To address challenges in analyzing high-dimensional blood gene expression data with limited samples.
• To identify critical genes for AD diagnosis and classification using advanced computational techniques.

Main Methods:

• Applied four feature selection methods (Chi-square, ANOVA, RFE, ElasticNet) to identify AD-related genes.
• Developed two deep learning models for AD classification based on selected gene biomarkers.
• Utilized nested five-fold cross-validation and SHapley Additive exPlanations (XAI) for model validation and gene ranking.
• Employed Generative Adversarial Network (GAN)-based data augmentation to enhance model generalization with small sample sizes.

Main Results:

• Successfully identified critical blood gene biomarkers for AD diagnosis.
• Achieved 91% accuracy and 95% precision in classifying AD samples using a deep neural network.
• Demonstrated significant improvements in precision and interpretability through feature selection and data augmentation.

Conclusions:

• Blood gene expression analysis, combined with advanced feature selection and deep learning, offers a viable strategy for early AD detection.
• The developed methodology enhances the accuracy and interpretability of diagnosing Alzheimer's disease.
• This research paves the way for more accessible and cost-effective early diagnostic tools for Alzheimer's disease.