Abstract Body (Do not enter title and authors here): Background: Heart Failure with preserved ejection fraction (HFpEF) is a heterogeneous syndrome with complex pathophysiology, limiting discovery in traditional genome-wide association studies (GWAS) as well as impeding development of targeted therapies. Hence, identifying distinct HFpEF subgroups is crucial for precision medicine. We aimed to delineate HFpEF subphenotypes using machine learning on electronic health record (EHR) data and uncover their genetic signals via targeted GWAS and systems biology approaches.
Hypothesis: We hypothesized that machine learning-derived HFpEF subphenotypes would exhibit unique genetic architectures, enabling discovery of novel, subphenotype-specific biological pathways.
Methods: From 55,916 HFpEF patients, we constructed a patient network from a 20,000-patient subset utilizing Iterative Random Forest - Leave One Out Prediction (iRF-LOOP). Iterative Random Forest (iRF) identified dominant EHR features, yielding six clinical subphenotypes. Cluster-specific GWAS (cases: cluster members; controls: non-HF patients) used a relaxed significance (1×10^-5). Genetic variants were mapped to genes via nearest gene, MAGMA, and H-MAGMA. Systems biology tools GRIN (Geneset Refinement using Interacting Networks) and MENTOR (Multiplex Embedding of Networks for Team-based Omics Research) functionally grouped genes within each cluster.
Results: We discovered six dominant HFpEF sub-phenotypes with diverse clinico-pathologic patterns, including cardiorenal, inflammatory/immune, and metabolic (e.g., younger, late-onset diabetes) phenotypes, plus groups defined by age/hypertension/atrial fibrillation or younger females with minimal comorbidities. Subphenotype-specific GWAS identified multiple genetic variants per cluster. Posterior systems biology analysis revealed distinct biological pathways (e.g., lipid metabolism, inflammation, fibrosis, calcium handling) aligning with observed clinical traits in the subphenotypes. This validated our approach and uncovered novel pathways, confirming unique genetic architectures influencing specific biological processes.
Conclusions: This integrated approach identified and characterized distinct genetic architectures for machine learning-derived HFpEF subphenotypes. Our findings advance the understanding of HFpEF heterogeneity by pinpointing specific genetic signals and pathways, providing a foundation for precision medicine in HFpEF.