Regularized Machine Learning Models for Prediction of Metabolic Syndrome Using GCKR, APOA5, and BUD13 Gene Variants:...

Area of Science:

• Genetics and Genomics
• Machine Learning
• Public Health

Background:

• Metabolic syndrome (MetS) presents a significant global health challenge due to its complex, multifactorial nature.
• Identifying individuals at high risk for MetS is crucial for effective prevention and management strategies.

Purpose of the Study:

• To classify Metabolic Syndrome (MetS) using regularized machine learning models.
• To evaluate the predictive power of genetic risk variants (GCKR, BUD13, APOA5) and environmental factors in MetS classification.
• To compare the performance of various regularization techniques against classical logistic regression.

Main Methods:

• A cohort study involving 2,346 cases and 2,203 controls from the Tehran Cardiometabolic Genetic Study (TCGS).
• Application of regularization methods including LASSO, Ridge Regression, Elastic Net, adaptive LASSO, and adaptive Elastic Net.
• Evaluation using 10-repeated 10-fold cross-validation with metrics like accuracy, AUC-ROC, and AUC-PR.

Main Results:

• Over 50% of participants developed MetS during the follow-up period.
• MetS was significantly associated with age, gender, BMI, schooling, and specific genetic risk variants.
• Regularized machine learning models demonstrated superior performance compared to logistic regression, with adaptive LASSO being the most parsimonious.

Conclusions:

• Regularized machine learning models offer enhanced accuracy and parsimony for MetS classification.
• These models can serve as a basis for clinical decision support tools.
• Integrating genetic and demographic data aids in identifying individuals at high risk for MetS.