Deciphering the Genetic Architecture of Heart Failure with Preserved Ejection Fraction - Project Summary Over the past two decades, advances in sequencing, statistical methods, and machine learning have transformed our understanding of the genetic basis of disease. Yet, progress across diseases has not been even: more than 300 loci have been discovered for coronary artery disease, but only two for heart failure with preserved ejection fraction (HFpEF). Given that there are no truly disease-modifying treatments and genomic insights triple drug discovery success, there are urgent calls to better understand HFpEF genetics. The goal of this proposal is to decipher the genetic architecture of HFpEF and identify novel therapeutic targets. Current limitations for HFpEF genomic discovery include imprecise phenotyping, lack of echocardiograms in biobanks, and parsimonious use of biological data to define HFpEF. Dr. O’Sullivan’s preliminary data begins to address these limitations. He developed a model trained to predict HFpEF from 40 biomarkers and deployed this model to the UK Biobank, assigning a predicted HFpEF probability to the ~500k participants. Using these probabilities as input for a genome-wide association study (GWAS), he identified over 50 HFpEF loci. This work demonstrates the feasibility of his machine learning approach and provides data for prioritizing causal loci. However, his current model lacks critical cardiac data: ECGs and imaging, and it remains unclear which of the identified loci represent novel therapeutic pathways. To address these unmet needs, Aim 1 will enhance his machine learning model by incorporating cardiac MRIs and ECGs. These cardiac-specific data will improve model specificity, facilitating enhanced genomic discovery. Aim 2 will redefine diastolic function assessment using vision encoders. These models will extract biological features from full diastolic echocardiogram images to generate novel diastolic scores for each patient, which will then be used as input for GWAS using linked genetic data. This will be the first effort to define the genomics of diastolic dysfunction, a hallmark of HFpEF. Aim 3 will identify causal gene-protein pathways dysregulated in HFpEF through proteomics and Mendelian Randomization. The proposed didactic and applied professional development experiences, including training in advanced machine learning and genomics, will facilitate Dr. O’Sullivan’s transition to scientific independence. His mentorship team, led by Dr. Euan Ashley, includes experts in heart failure (Lewis, advisor), machine learning (Chaudhari and Pirruccello, advisors), computer science (Rivas, mentor and Chaudhari, advisor), and statistical genomics (Assimes and Rivas, advisors). The research and training plan proposed in this K08 application will develop Dr. O’Sullivan into a unique and highly skilled physician-scientist, ready to compete for R-level funding and launch his independent career in computational genomics research.
