Microbiome Atherosclerosis Risk Simulator (MARS)

Atherosclerotic Cardiovascular Disease (ASCVD) is the leading cause of death in the United States. This inflammatory disease is characterized by fatty plaque buildup in the arteries, leading to inflammation. In recent years, research has begun to support the role of gut dysbiosis in heart disease. This project aimed to develop a machine learning model, MARS, that could preemptively predict a patient’s risk for ASCVD through analyzing their gut composition. Metagenomic data from ASCVD and healthy groups was filtered for 391 unique relevant bacterial species. MARS was trained with 80% of the data and tested with the remaining 20%. Class-balancing and dimensionality reduction optimized the model to only take an input of the top 50 microbes for efficiency. The statistical significance of the selected features was verified using a Mann-Whitney U test. Additionally, the model’s predictive threshold was tuned down to 30% from 50% to minimize false negatives. Finally, a diet simulation algorithm was built into a web application to model the impact of Mediterranean and High-Choline diets on ASCVD risk. The optimized model received a ROC-AUC score of 0.87, indicating a strong capability to differentiate between healthy and ASCVD cohorts. Statistical analysis confirmed the top feature, Bacteroides vulgatus (p = 3.95 ∙ 10⁻¹²), was inversely elevated in the diseased group. Despite being a protective species, Bacteroides vulgatus were elevated in patients with ASCVD, suggesting a compensatory response to arterial inflammation. MARS could serve as a powerful and cheap diagnostic tool for clinicians with more fine-tuning and accuracy in the future.