Predicting Cardiovascular Disease Risk Through Machine Learning Analysis of Gut Microbiota Metabolite Flux Modeling with Flux-Balance Analysis

Cardiovascular disease (CVD) is the leading cause of death globally, necessitating more accessible and cost-effective diagnostic methods. The gut microbiome has emerged as a promising avenue for CVD diagnosis, as its bacterial metabolites influence cardiovascular health. This study proposes a novel machine learning (ML) approach that leverages metabolite flux data obtained through flux-balance analysis (FBA) to predict CVD. Using stool sample data from the American Gut Project (AGP), bacterial relative abundance was first used to construct personalized metabolic models for each patient. FBA was applied to these models to estimate metabolite fluxes, which were then used to train various ML models, including Random Forest (RF), Support Vector Machine (SVM), Elastic Net (EN), Decision Tree (DT), and Neural Networks (NN). Initial results showed that models trained on bacterial relative abundance alone were ineffective (AUC ~0.70). However, models trained on metabolite fluxes significantly improved accuracy (top AUC ~0.76). Further enhancement was achieved by integrating a gene expression prediction layer, addressing a key limitation of traditional FBA, which does not consider gene activity levels. The inclusion of gene expression adjustments led to the highest model performance. These findings suggest that metabolite-based ML models can provide a more accurate and physiologically relevant method for CVD diagnosis than bacterial abundance alone. Future work should expand gene expression modeling beyond a single metabolite to enhance prediction accuracy. This study demonstrates the potential of integrating metabolic modeling and ML for non-invasive disease diagnosis, paving the way for gut microbiome-informed clinical applications.