LD-informed deep learning for Alzheimer's gene loci detection using WGS data.

Area of Science:

• Genomics
• Artificial Intelligence
• Neuroscience

Background:

• Genomic datasets are growing exponentially, requiring advanced tools for genetic loci identification.
• Alzheimer's disease (AD) research needs efficient methods to analyze large-scale, high-throughput sequencing data.
• Identifying genetic factors for AD is crucial for understanding disease mechanisms and developing targeted therapies.

Purpose of the Study:

• To introduce Deep-Block, a multi-stage deep learning framework for identifying genetic regions associated with Alzheimer's disease.
• To incorporate biological knowledge into an AI architecture for enhanced genetic analysis.
• To identify novel single nucleotide polymorphisms (SNPs) contributing to AD risk.

Main Methods:

• Genome segmentation using linkage disequilibrium (LD) patterns.
• Sparse attention mechanisms for selecting relevant LD blocks.
• TabNet and Random Forest algorithms for SNP feature importance quantification.
• Application to a large whole genome sequencing (WGS) dataset from the Alzheimer's Disease Sequencing Project (ADSP).

Main Results:

• Identification and ranking of 30,218 LD blocks based on AD relevance.
• Discovery of novel SNPs within the top 1,500 LD blocks, alongside confirmation of known variants like APOE rs429358.
• Functional evidence for identified variants through expression Quantitative Trait Loci (eQTL) analysis across 13 brain regions.
• Cross-validation against established AD loci from the European Alzheimer's and Dementia Biobank (EADB) and GWAS catalog.

Conclusions:

• Deep-Block effectively processes large-scale sequencing data, preserving SNP interactions and minimizing information loss.
• Tissue-specific eQTL analysis supports the functional relevance of identified variants in brain regions.
• The framework successfully identified known and novel genetic variants, advancing the understanding of AD's genetic architecture.