Abstrat:　 Genomic sequences of viral origin are widespread in eukaryotic genomes, yet contributions from double-stranded (ds)DNA viruses remain poorly characterized. Here, we present a bioinformatic framework for identifying dsDNA viral regions (VRs) embedded within eukaryotic genome assemblies. Applying this framework to over 37,000 eukaryotic genomes, we detected more than 780,000 endogenous VRs across diverse host lineages. Our analyses enable taxonomic assignment and phylogenetic reconstruction of endogenous viral elements, expanding the eukaryotes-virus association network. Together, this framework provides a computational foundation for exploring endogenous dsDNA viral diversity at genome scale and for investigating long-term virus–host coevolution.

Abstract:　Drug discovery is inherently time‑consuming and labor‑intensive, creating a strong demand for innovative methods that can accelerate the identification of novel therapeutic molecules. In recent years, deep generative models for molecular generation have gained traction as a promising way to streamline early‑stage drug discovery. However, most existing methods are chemistry‑centric approaches that rely solely on chemical structures and their associated properties, limiting their ability to capture biological context. To address this limitation, we have developed molecular generation methods that incorporate transcriptome profiles, thereby enabling drug candidate design that leverages more comprehensive biological information. In this presentation, we present our work on both de novo and scaffold‑constrained molecular generation informed by transcriptomic profiles. Furthermore, we highlight a case study where the integration of our generative models with experimental screening facilitated the discovery of novel inhibitors.