Explainable variational autoencoder (E-VAE) model using genome-wide SNPs to predict dementia.

Area of Science:

• Neuroscience
• Genetics
• Artificial Intelligence

Background:

• Alzheimer's disease (AD) and related dementias (ADRD) are complex neurodegenerative conditions.
• Genome-wide association studies (GWAS) provide extensive genetic variant data linked to ADRD.
• Deep learning offers a powerful method for analyzing large GWAS datasets to uncover biological mechanisms.

Purpose of the Study:

• To develop and validate a deep learning model for classifying ADRD using genetic data.
• To explore the potential of deep learning in elucidating ADRD biological mechanisms.
• To assess the generalizability of the developed model in an independent cohort.

Main Methods:

• Developed an explainable variational autoencoder (E-VAE) classifier using GWAS SNP data from 2714 participants in the Health and Retirement Study (HRS).
• Validated the E-VAE model's generalizability using data from 234 participants in the Religious Orders Study and Memory and Aging Project (ROSMAP).
• Employed a linear decoder approach to extract weights for biological interpretation of latent features.

Main Results:

• The E-VAE model achieved a predictive accuracy of 0.71 (AUC 0.69) in the HRS test dataset.
• The model demonstrated generalizability with an accuracy of 0.62 (AUC 0.63) in the independent ROSMAP dataset.
• Identified latent features through E-VAE for potential biological interpretation.

Conclusions:

• This study is the first to demonstrate the generalizability of a deep learning prediction model for dementia using genetic variants in an independent cohort.
• The identified latent features from E-VAE can contribute to understanding AD/ADRD biology.
• The E-VAE approach shows potential for improved characterization of dementia disease status.