ExAD-GNN: Explainable Graph Neural Network for Alzheimer's Disease State Prediction from Single-cell Data

Area of Science:

• Computational biology
• Neuroscience
• Genomics

Background:

• Alzheimer's disease (AD) diagnosis requires early detection for effective treatment.
• Current machine learning methods for AD diagnosis lack molecular detail and ignore brain heterogeneity.
• This limits understanding of complex disease mechanisms.

Purpose of the Study:

• To develop an Explainable Graph Neural Network (ExAD-GNN) for predicting AD from single-cell sequencing data.
• To achieve cellular-level AD pathology prediction and identify cell-type-specific AD marker genes.
• To provide molecular insights into AD pathology using interpretable AI.

Main Methods:

• Utilized K Nearest Neighbours (KNN) graphs from single-cell expression profiles.
• Developed ExAD-GNN, an Explainable Graph Neural Network model.
• Employed a learnable gene importance metric for marker gene identification.

Main Results:

• ExAD-GNN demonstrated improved accuracy and robustness in AD prediction across diverse cell types and samples compared to state-of-the-art methods.
• The model successfully identified key AD risk genes, validated by literature.
• The explainability scheme effectively highlighted important biological markers.

Conclusions:

• ExAD-GNN offers a powerful, interpretable approach for AD diagnosis using single-cell RNA sequencing data.
• The tool enhances understanding of AD at a molecular and cellular level.
• ExAD-GNN is publicly available to aid scientific research in neurodegenerative disorders.