Insights into heart failure metabolite markers through explainable machine learning.

Area of Science:

• Cardiovascular Research
• Metabolomics
• Artificial Intelligence in Medicine

Background:

• Metabolomics provides insights into molecular traits for personalized cardiovascular strategies.
• Heart failure (HF) diagnosis and treatment can be improved by understanding molecular signatures.
• Machine learning (ML) and explainable artificial intelligence (XAI) offer powerful tools for analyzing complex biological data.

Purpose of the Study:

• To apply ML and XAI algorithms to identify discriminant molecular signatures in heart failure (HF).
• To uncover specific metabolites with significant predictive value for HF using targeted metabolomics data.
• To validate the utility of ML and XAI in advancing personalized cardiovascular healthcare.

Main Methods:

• Analysis of 55 metabolites from 124 plasma samples (53 HF patients, 71 controls).
• Comparison of Ridge Logistic Regression, Support Vector Machine, and eXtreme Gradient Boosting models.
• Application of permutation-based variable importance, Local Interpretable Model-agnostic Explanations (LIME), and H-Friedman statistics for model interpretability and interaction analysis.

Main Results:

• All ML models achieved high prediction accuracy for HF: Ridge Logistic Regression (84.0%), Support Vector Machine (85.7%), and eXtreme Gradient Boosting (84.8%).
• Established HF-associated metabolites (glucose, cholesterol, C18:1 carnitine) were reaffirmed.
• Novel discriminator lignoceric acid (C24:0 fatty acid) was identified and validated in a replication cohort, alongside significant metabolite interactions.

Conclusions:

• ML and XAI effectively identify key metabolites and their interactions for accurate HF prediction.
• Lignoceric acid shows potential as a novel metabolite biomarker for heart failure.
• The study demonstrates the power of integrating ML and XAI in metabolomics for personalized cardiovascular medicine.