Mapping Alzheimer's disease heterogeneity through exploratory unsupervised learning

Area of Science:

• Neuroscience
• Genetics
• Computational Biology

Background:

• Alzheimer's disease (AD) affects millions globally, with diagnosis challenged by long preclinical phases and heterogeneity.
• Current AD research often uses data lacking diagnostic quality, potentially skewing results.
• Biomarker confirmation is crucial for accurate AD diagnosis and research.

Purpose of the Study:

• To explore if curated genetic markers can reveal substructure within sporadic AD (sAD) using unsupervised machine learning.
• To identify genetically distinguishable subgroups within sAD.
• To enhance the quality of genetic data used in AD subtype research.

Main Methods:

• Utilized biomarker-confirmed samples from the Alzheimer's Disease Neuroimaging Initiative (ADNI).
• Selected two sets of single nucleotide polymorphisms (SNPs): candidate SNPs and those from genome-wide association studies (GWAS).
• Applied unsupervised machine learning clustering, with agglomerative hierarchical clustering showing robust performance.

Main Results:

• Clustering analysis revealed a reproducible binary structure within sAD samples across different SNP sets.
• This suggests the presence of genetically distinct subgroups within the sAD population.
• Agglomerative hierarchical clustering demonstrated consistent and robust performance.

Conclusions:

• Findings support the biological heterogeneity of AD.
• Curated SNP panels and unsupervised learning can uncover meaningful substructure in sAD.
• Integrating high-quality genetic data is valuable for AD subtype research.