An Interpretable Deep Learning Approach for Alzheimer's Disease Diagnosis Using Gene Expression Data.

Area of Science:

• Computational biology
• Neuroscience
• Machine learning

Background:

• Global population aging increases Alzheimer's disease (AD) diagnostic urgency.
• Gene expression analysis is cost-effective but faces high dimensionality challenges in AD diagnosis.
• The curse of dimensionality hinders accurate AD diagnosis from gene expression data.

Purpose of the Study:

• To develop a novel, interpretable deep learning approach for accurate Alzheimer's disease diagnosis.
• To address the challenges of high dimensionality and small sample sizes in gene expression data for AD.
• To enhance model efficiency and provide biological insights into AD mechanisms.

Main Methods:

• Utilized a shallow sparse autoencoder for dimensionality reduction.
• Combined autoencoder with XGBoost classifier for AD diagnosis.
• Developed a dynamic feature selection algorithm for improved efficiency.

Main Results:

• Achieved a high Area Under the Receiver Operating Characteristic curve (AUROC) of 95.13%.
• Demonstrated strong generalization performance across multiple public datasets (ADNI, ANM1, ANM2).
• Provided biological interpretability via enrichment analysis, identifying potential therapeutic targets.

Conclusions:

• The proposed interpretable deep learning method is effective for early and accurate Alzheimer's disease diagnosis.
• The approach offers a promising tool for clinical application, overcoming common challenges in gene expression analysis.
• Biological insights derived from the model can advance understanding of AD mechanisms and therapeutic strategies.