Machine Learning Methods for Classifying Multiple Sclerosis and Alzheimer's Disease Using Genomic Data.

Area of Science:

• Genomics
• Computational Biology
• Biostatistics

Background:

• Complex diseases are challenging to predict due to their polygenic and multifactorial nature.
• Genomic data analysis holds potential for understanding disease predisposition.
• Existing methods like polygenic risk scores (PRS) have limitations in capturing complex genetic patterns.

Purpose of the Study:

• To evaluate machine learning (ML) models for predicting genomic predisposition to complex diseases using UK Biobank data.
• To compare the performance of logistic regression (LR), ensemble tree methods, and deep learning models.
• To investigate the utility of ML in identifying key genomic variants contributing to disease risk.

Main Methods:

• Analysis of genomic data from the UK Biobank.
• Implementation and comparison of logistic regression, ensemble tree methods, and deep learning models.
• Application of explainability tools to interpret ML model predictions for multiple sclerosis (MS).

Main Results:

• Logistic regression demonstrated superior stability and performance compared to deep learning models across data subsets.
• ML models maintained performance despite correlated genomic features arising from linkage disequilibrium.
• Polygenic risk scores (PRS) performed at an average level, generally below ML methods.
• Explainability analysis for MS identified non-coding variants, particularly expression/splicing quantitative trait loci near immune-related genes (including HLA), confirming polygenicity.

Conclusions:

• Machine learning, particularly logistic regression, offers a robust approach for predicting genomic predisposition to complex diseases.
• ML models can effectively handle correlated genomic data and provide insights into disease-associated variants.
• Further development of ML techniques is crucial for advancing predictive modeling in complex disease genomics.