ABSTRACT Early-stage heart failure (HF), defined broadly as asymptomatic structural heart disease, is a growing public health concern and has been receiving greater attention in clinical settings. Early-stage HF is a risk factor for advanced HF, but there is heterogeneity in this risk. While there are known biomarkers and biological mechanisms of advanced HF, little is known about the potential entity of early-stage HF that is distinct from underlying HF risk factors. Additionally, individual -omics platforms have been applied to discover novel biology and biomarkers of advanced HF, but multi-omics data integration methods have not yet been applied in this space. Through preliminary analyses, I have discovered novel proteins for early-stage HF that are distinct from underlying risk factors and advanced HF. With these initial results, I hypothesize that molecular features of cardiac inflammation are significant biomarkers for early-stage HF, and that we can optimize biomarker and biological mechanism discovery by applying integrative -omics techniques. The broad objective of this proposal is to discover novel biomarkers and biological mechanisms underpinning early-stage HF that have prognostic capabilities for advanced HF. Understanding how early-stage HF is distinct from underlying risk factors and advanced HF at a molecular and clinical level could enable the development of novel diagnostic tests and preemptive treatment measures to forestall adverse late-stage outcomes. I propose to accomplish this by utilizing state-of-the-art machine learning methodologies for the analysis of multi-omics and clinical data from large-scale population cohorts. First, I will develop and validate an automated computable phenotype for the identification of early-stage HF using clinical data and imaging-derived measures of cardiac structure and function. I will then apply this computable phenotype in the UK Biobank to identify significant protein and metabolite features of predicted early-stage HF cases using separate single-omics analyses. After identifying candidate biomarkers, I will assess their causality and biological relevance by performing Mendelian Randomization and pathway enrichment analysis. To test the impact of data integration on model predictive ability, I will apply a modified version of MiNet, an interpretable pathway-associated deep neural network for diseas prediction. Using MiNet, I will integrate proteomics and metabolomics data to predict Stage B HF cases versus Stage A HF controls and compare the neural network’s performance to those from the single-omics models. From the trained deep learning model, I will compare pathway node activation in the neural network to the single-omics pathway analysis results to assess whether multi-omics integration affects the richness of biological findings from a model. With this work, I will improve knowledge of early-stage heart failure biomarkers and biology; I will also expand upon the application of evol...