Machine learning approaches for biomarker discovery to predict large-artery atherosclerosis

Area of Science:

• Biomedical Informatics
• Cardiovascular Research
• Metabolomics

Background:

• Large-artery atherosclerosis (LAA) is a primary cause of cerebrovascular disease, but its diagnosis is challenging and resource-intensive.
• Existing biomarkers for LAA prediction show inconsistent results.
• Metabolites are increasingly recognized as potential biomarkers for various health conditions.

Purpose of the Study:

• To develop and validate a novel machine learning-based method for predicting LAA.
• To identify reliable clinical and metabolic biomarkers for LAA prediction.
• To improve the efficiency and accuracy of LAA diagnosis.

Main Methods:

• Integration of multiple machine learning algorithms with feature selection techniques to analyze multidimensional data.
• Utilized logistic regression (LR) as the top-performing predictive model.
• Employed recursive feature elimination and cross-validation for robust biomarker identification.

Main Results:

• The logistic regression model achieved an AUC of 0.92 with 62 features in external validation.
• Key predictors included body mass index, smoking, diabetes, hypertension, hyperlipidemia medications, and metabolites from aminoacyl-tRNA biosynthesis and lipid metabolism.
• A refined set of 27 shared features improved the LR model's AUC to 0.93.

Conclusions:

• Machine learning combined with feature selection effectively identifies biomarkers for LAA.
• Shared features across models enhance prediction reliability and are valuable for future LAA biomarker discovery.
• This approach offers a promising, data-driven strategy for LAA prediction.