Deep Learning Integration of Multi-Species Functional Genomics to Reveal Conserved Gene Regulatory Logic

Abstract

Understanding conserved gene regulatory logic across species is essential for unraveling the mechanisms controlling gene expression and cellular states. Here, we present a deep learning–based framework that integrates multi-species functional genomics datasets, including sequence, epigenomic, and perturbation data, to identify and characterize conserved regulatory elements and networks across plants, fungi, and metazoans. By leveraging high-quality, harmonized datasets and multi-modal architectures, our approach captures both sequence- and chromatin-based regulatory features and enables cross-species prediction of gene regulatory activity. We demonstrate that regulatory motifs, chromatin states, and meta-gene expression patterns exhibit substantial conservation, even across evolutionarily distant species. Our results highlight the potential of self-supervised deep learning to uncover subtle and complex regulatory grammar, advancing the understanding of evolutionary conservation, functional genomics, and gene regulatory network engineering. This framework provides a scalable, reproducible, and open-science approach to studying gene regulation across diverse taxa