HF-ETIOLOGY: Heart Failure Endotypes from Ethical, Multi-Modal AI driven and Molecular/Phenotypic Data Integration Enabled Discovery

Abstract Precision medicine is rapidly evolving through integrating multi-modal data to tailor treatments and deepen our understanding of diseases like heart failure (HF). Our initiative, HF-ETIOLOGY develops an ethical multi-modal AI framework that harnesses the power of multi-modal data—phenotypic, multi-omic, and socio-behavioral—to identify distinct HF endotypes. Our approach is distinguished by its co-design and iterative development of novel multi-modal AI models for disease endotyping, including advanced Bayesian generative tensor models and temporal tensor models and network medicine approaches. These methods allow us to integrate diverse data modalities and structures, such as longitudinal clinical data and complex multi-omic networks, seamlessly with behavioral and Social Determinants of Health (SDoH) factors. Our project centers around four main objectives: 1) Establish a FAIR-CARE framework to co-design HF- ETIOLOGY data and models; 2) Co-design Generative and Adjustable Prior Bayesian Tensor Factorization (GAP-BTF) to integrate behavioral and SDoH factors with multi-omic network feature learning to identify HF endotypes; 3) Co-design temporal non-negative tensor factorization model (TNTF) to integrate longitudinal phenotypic data while jointly modeling behavioral and SDoH factors for HF endotyping; 4) Identify likely risk genes/targets and pathways and investigate drug repurposing by incorporating SDoH factors for clinically relevant HF endotypes with MAI and network medicine co-design. Central to our methodology is a co-design framework that involves continuous engagement with stakeholders. This collaborative approach ensures that the development of our AI models is informed by clinical insights and aligned with ethical standards, thereby enhancing the practicality and relevance of our research in the clinical setting.