Large Scale Artificial Intelligence Models Unifying Epigenomic and Transcriptomic Signals to Interpret Human Disease Genetics

Abstract

Recent advances in high-throughput sequencing and large-scale international consortia have generated extensive epigenomic and transcriptomic datasets across diverse human tissues and disease contexts. These resources offer unprecedented opportunities to interpret the functional consequences of genetic variation, particularly for non-coding variants that account for the majority of heritability in common human diseases. However, traditional analytical approaches often model epigenomic and transcriptomic signals independently, limiting their ability to capture the complex, context-dependent regulatory mechanisms that connect genotype to phenotype. This work highlights the rationale and emerging methodologies for integrative modeling of epigenomic landscapes and transcriptomic profiles using large-scale artificial intelligence (AI) frameworks. By jointly learning from complementary molecular modalities, such models can uncover shared and modality-specific regulatory representations, align chromatin states with gene expression outputs, and better characterize disease-relevant regulatory mechanisms. We discuss data integration strategies, representation learning techniques, and multimodal alignment approaches that enable unified interpretation of epigenomic and transcriptomic signals at scale. Integrative AI-driven frameworks hold substantial promise for improving variant prioritization, elucidating gene regulatory mechanisms, and advancing precision medicine through more accurate interpretation of human disease genetics.