Basic Science and Pathogenesis.

Area of Science:

• Genetics
• Artificial Intelligence
• Data Science

Background:

• Alzheimer's Disease (AD) is a leading cause of cognitive decline, with early diagnosis challenging due to complex pathogenesis and symptom overlap with aging.
• Genetic phenotyping aids in identifying at-risk individuals, but current Polygenic Risk Scores (PRS) do not fully capture AD's genetic complexity.

Purpose of the Study:

• To enhance the predictive accuracy of Alzheimer's Disease using advanced AI and Data Science techniques.
• To reduce computational time in genetic prediction models for AD.
• To identify novel genetic variants associated with AD.

Main Methods:

• Utilized a dataset of 534 individuals (AD and healthy controls) with 297,678 genetic variants.
• Implemented a novel one-hot-encoding framework for transposed data enabling distributed processing and efficiency.
• Applied filter-based feature selection to reduce genetic variants from 297,678 to 2,939.
• Employed distributed machine learning models (Logistic Regression, SVM) validated via 10-fold cross-validation and AUROC metrics.

Main Results:

• The developed global SVM model achieved an Area Under the Receiver Operating Characteristic curve (AUROC) of 0.94, significantly outperforming existing PRS models (AUROC 0.71).
• Identified 38 novel genetic variants associated with AD, contributing to a broader understanding of AD pathogenesis.
• Enhanced predictive accuracy facilitates earlier AD detection and personalized healthcare.

Conclusions:

• An AI-driven approach effectively addresses the limitations of PRS in capturing AD's genetic complexity.
• The methodology improves feature selection, computational efficiency, and predictive accuracy for AD genetic prediction.
• Future work includes extending the study with larger datasets and enhanced feature engineering for more robust predictions.