Identification of Key Genomic Variants for The Early Detection of Alzheimer’s Disease: An Explainable Approach

Alzheimer’s disease (AD) is the most common form of degenerative dementia. It can be considered a complex disease, characterized by the combination of several genomic variants and genes, and increasing knowledge of the causes related to AD can promote its early diagnosis. In this study, we propose an explainable machine learning approach to classifying AD patients and finding which mutations could be a key factor in the potential evolution of the pathology. For this purpose, we used some longitudinal studies collected in the ADNI repository. Genomic variants from the ADNI repository have been fed to different machine learning algorithms and evaluation measures to quantify the performance of classifiers, and finally, the SHAP explanation method was used to highlight the most significant genetic variants potentially linked to the pathology evolution or with a protective role. The analysis results show that the most significant genomic variations for classifying AD patients are on chromosomes 19, but also on 9, 13, 14 and 17.