Polygenic risk scores outperform machine learning methods in predicting coronary artery disease status

Area of Science:

• Cardiovascular Genetics
• Computational Biology
• Medical Informatics

Background:

• Coronary artery disease (CAD) is a leading global cause of death with significant genetic influence.
• Existing polygenic risk scores (PRS) may not capture nonlinear genetic effects or all relevant genetic loci.
• Advanced algorithms are needed for accurate CAD risk prediction.

Purpose of the Study:

• To benchmark various algorithms, including PRS and machine learning methods, for classifying CAD status.
• To identify the most accurate method for predicting CAD risk in a German population.
• To compare the performance of PRS against traditional and machine learning approaches.

Main Methods:

• A dataset of 7,736 CAD cases and 6,774 controls from Germany was used for model training.
• Algorithms benchmarked included PRS, logistic regression, Naïve Bayes, Random Forests, SVM, and Gradient Boosting.
• Models were validated on an independent German dataset of 527 CAD cases and 473 controls.

Main Results:

• Polygenic risk scores (PRS) achieved the highest accuracy, with an Area Under the Receiver Operating Curve (AUC) of 0.92 in the test data.
• Naïve Bayes and Support Vector Machines showed moderate performance (AUC ~0.81).
• Random Forests and Gradient Boosting yielded lower AUC values (~0.75).

Conclusions:

• Polygenic risk scores (PRS) demonstrate superior performance in predicting CAD compared to tested machine learning algorithms.
• PRS offers a more effective approach for CAD risk stratification.
• Further research may explore integrating nonlinear effects into PRS models.