Machine learning-based quantification for disease uncertainty increases the statistical power of genetic association...

Area of Science:

• Genetics
• Neuroscience
• Computational Biology

Background:

• Genome-wide association studies (GWAS) require large sample sizes to identify genetic variants associated with Alzheimer's disease (AD).
• Traditional GWAS methods often exclude individuals with mild cognitive impairment (MCI) or unknown cognitive status, potentially limiting power.

Purpose of the Study:

• To develop and validate a novel method for Alzheimer's disease (AD) genome-wide association studies (GWAS) that incorporates individuals with mild cognitive impairment (MCI) and unknown cognitive status.
• To improve the statistical power of AD GWAS by utilizing a machine learning-based prediction model.

Main Methods:

• Developed a machine learning-based AD prediction model to assign phenotypes to individuals with MCI and unknown cognitive status.
• Employed a weighting imputed phenotypes method and penalized logistic regression for GWAS analysis.
• Validated the method through simulation analyses and application to real-world data.

Main Results:

• The weighting imputed phenotypes method demonstrated increased statistical power compared to ordinary logistic regression using only AD cases and controls.
• The penalized logistic method achieved a high Area Under the Curve (AUC) of 0.96 for AD prediction in real-world data.
• Identified significant associations (P<5.0×10-8) of AD with variants in the APOE region and a novel association with rs143625563 in LMX1A.

Conclusions:

• The developed method effectively incorporates individuals with MCI and unknown cognitive status into AD GWAS, thereby enhancing statistical power.
• The study identified a novel genetic association with the LMX1A gene, contributing to our understanding of AD pathogenesis.
• The simulation codes are publicly available for further research.