Identification of Multiomic Interactions Directing Mammalian Heart Development

Congenital heart disorders are a leading cause of morbidity and mortality in children, affecting an estimated 1 in 100 live births each year. These conditions often result from genetic alterations in the noncoding genome that disrupt the normal development of the heart, leading to malformations and functional impairments. Intricate epigenetic and transcriptomic dynamics govern these processes of cellular differentiation and organogenesis. However, even with the newest technologies in single-cell sequencing, it remains difficult and resource-intensive for researchers to quantify the regulatory behavior between chromatin accessibility, gene expression, and ultimately downstream changes in cell identity. Current methods to analyze single-cell sequencing data rely on mathematical simplifications for computational expediency that may not fully mirror biological reality. This study aims to advance the understanding of cardiac development by creating new AI-based computational tools to profile the epigenetic and transcriptomic landscapes simultaneously. Our novel transformer architecture achieves a high predictive accuracy by analyzing multiomic sequencing data from mouse hearts at three stages of embryonic development. Attention‐based linkage analysis not only recapitulates known temporally restricted cell type-specific regulatory interactions but also uncovers novel candidate epigenetic drivers of cardiomyocyte lineage commitment. These findings demonstrate the power of interpretable transformer models to computationally reconstruct dynamic regulatory networks, which would aid future research in elucidating the exact mechanisms of cell fate determination and pinpointing the etiologies of congenital disorders.