From:  AI-driven drug discovery and repurposing using multi-omics for myocardial infarction and heart failure

 Summary of representative XAI applications in cardiovascular research, highlighting the data types used, AI methods, and performance outcomes

Study (year)Data & cohortAI methodXAI methodTask/OutcomePerformance
Adams et al. [48] (2020)Genomic (GWAS of statin-treated patients, n = 5,890)Random forest (RF-IFRS)Decision tree-based network visualizationPredict risk of on-statin MACENo standard metric reported (identified 6 variant networks; highest OR ~4.5 for the high-risk subgroup)
Pezoulas et al. [49] (2022)Metabolomics (Young Finns cohort)XGBoost (hybrid ML model)SHAP (global feature importance)Predict carotid artery disease severity (intima-media thickness)High accuracy and sensitivity (~90% accuracy reported)
Yilmaz [50] (2023)Metabolomics (serum metabolites in acute MI patients vs. controls)GBTLIME (local explanation)Classify acute MI vs. healthy controlsHigh classification accuracy (effective separation of MI vs. control samples)
DeGroat et al. [51] (2024)Multi-omics (mRNA expression + SNPs; CVD patients vs. controls)XGBoost (with Bayesian tuning)SHAP (feature importance)Classify CVD patients vs. controls; identify key biomarkers100% accuracy on the test set (n = 15; AUC = 1.00)

AI: artificial intelligence; XAI: explainable AI; GWAS: genome-wide association studies; RF-IFRS: Random Forest Iterative Feature Reduction and Selection; MACE: major adverse cardiovascular events; XGBoost: eXtreme Gradient Boosting; ML: machine learning; SHAP: SHapley Additive exPlanations; MI: myocardial infarction; GBT: Gradient Boosted Trees; LIME: Local Interpretable Model-agnostic Explanations; SNPs: single nucleotide polymorphisms; CVD: cardiovascular disease; OR: odds ratio; AUC: area under the curve