Classification and deep-learning-based prediction of Alzheimer disease subtypes by using genomic data

Area of Science:

• Neuroscience
• Genetics
• Gerontology

Background:

• Late-onset Alzheimer's disease (LOAD) is a common, multifactorial neurodegenerative disorder in the elderly.
• LOAD presents with heterogeneous symptoms, and its genetic underpinnings, particularly for subtypes, remain incompletely understood.
• Previous genome-wide association studies (GWAS) identified LOAD risk factors but did not differentiate between LOAD subtypes.

Purpose of the Study:

• To investigate the genetic architecture of LOAD subtypes using Japanese GWAS data.
• To identify distinct genetic profiles associated with different LOAD patient groups.
• To explore potential links between identified genetic factors and physiological markers.

Main Methods:

• Analysis of Japanese GWAS data from a discovery cohort (1947 LOAD patients, 2192 controls) and a validation cohort (847 LOAD patients, 2298 controls).
• Identification of genetic risk factors and gene sets associated with distinct LOAD patient groups.
• Correlation analysis between genetic findings and routine blood test markers (albumin, hemoglobin).
• Development of a deep neural network model for predicting LOAD subtypes.

Main Results:

• Two distinct genetic subtypes of LOAD were identified.
• Subtype 1 was characterized by major LOAD risk genes (APOC1, APOC1P1) and immune-related genes (RELB, CBLC).
• Subtype 2 was characterized by genes associated with kidney disorders (AXDND1, FBP1, MIR2278).
• Analysis suggested a potential link between impaired kidney function and LOAD pathogenesis.
• A deep neural network model achieved prediction accuracies of 0.694 (discovery) and 0.687 (validation) for LOAD subtypes.

Conclusions:

• LOAD exhibits distinct genetic subtypes with potentially different pathogenic mechanisms.
• Genetic factors associated with kidney disorders may play a role in a subset of LOAD cases.
• These findings offer novel insights into LOAD pathogenesis and suggest avenues for subtype-specific diagnostics and therapeutics.