Predicting Alzheimer’s Disease Risk Based on Genetic Variant Analysis

Abstract

Alzheimer’s disease (AD) is a progressive neurodegenerative disorder and a leading cause of dementia worldwide. Early identification of individuals at high risk is crucial for timely intervention and improved patient outcomes. Investigating genetic variations, like mutations, for early disease prediction has become increasingly viable. Recent advances in genomics have highlighted the role of single nucleotide polymorphisms (SNPs) in the genetic predisposition to AD. Machine learning (ML) enables precise, early prediction of AD risk by decoding the complex genetic information from SNPs and biomarker patterns, which has a great potential to accelerate personalised interventions and therapeutic development. This thesis explores computational approaches for predicting susceptibility to AD using SNPs combined with genetic data from Genome-Wide Association Studies (GWAS). The study integrates SNP data from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) and the NHGRI-EBI GWAS Catalog, focusing on the most common overlapping SNPs between both datasets. An end-to-end pipeline was developed for SNP preprocessing and frequency-based feature extraction. Several ML models were tested, including Logistic Regression, Random Forest, Support Vector Machines, and ensemble classifiers. Rather than relying on default configurations, each method was carefully tuned using custom parameters to achieve optimal performance. Results demonstrate that certain ML models, particularly those using customtuned parameters, yield competitive accuracy in classifying AD risk. However, limitations such as imbalanced data and lack of deep feature extraction were noted. This lays the foundation for future work incorporating deep learning, advanced feature engineering, and multimodal data sources.