Evaluation of Machine Learning Methods to Predict Coronary Artery Disease Using Metabolomic Data

Area of Science:

• Biochemistry
• Computational Biology
• Cardiovascular Medicine

Background:

• Metabolomic data offers potential for non-invasive, cost-effective coronary artery disease (CAD) prediction.
• Traditional regression models may not fully capture complex metabolite interactions, limiting prediction accuracy.
• Supervised machine learning (ML) can leverage the high dimensionality of metabolomic data.

Purpose of the Study:

• To systematically implement and evaluate supervised ML methods for CAD prediction using metabolomic data.
• To compare the performance of ML approaches against traditional regression-based methods.
• To assess the potential of ML in fully exploiting metabolomic data richness for disease prediction.

Main Methods:

• Implementation and evaluation of L1 regression for CAD prediction.
• Application and assessment of random forest classifiers for CAD prediction.
• Comparative analysis of ML methods versus traditional regression approaches using metabolomic datasets.

Main Results:

• Supervised ML methods, including L1 regression and random forest, demonstrate potential for accurate CAD prediction.
• ML approaches show promise in overcoming limitations of traditional regression models in handling complex metabolite interactions.
• The study provides a systematic evaluation of different analytical strategies for metabolomic data in disease prediction.

Conclusions:

• Supervised ML methods offer a powerful alternative to traditional regression for metabolomic data analysis in CAD prediction.
• ML techniques can better account for metabolite interactions, potentially leading to improved diagnostic accuracy.
• Further research into ML applications can enhance non-invasive prediction of cardiovascular diseases.