Characterizing highly conserved fragments in 3′UTRs via statistical and transfer learning approaches

This is the authors' abstract. We don't add key points for this paper.

3' Untranslated regions (3'UTRs) serve as regulatory platforms that modulate steps in the central dogma through the binding of RNA-binding proteins and miRNAs. Their binding sites are often identified through orthologous regions among species. A separate but related discovery was the ultraconserved elements (UCEs) detected in human, rat, and mouse genomes two decades ago. However, knowledge about their functions is limited. Perplexingly, mutagenesis of UCEs produced no observable phenotypic differences. The majority of UCEs are non-coding, though ∼8% are located in the 3'UTRs. Given the importance of 3'UTRs in gene regulation, we use a computational approach to identify highly conserved fragments (CFs) that exhibit ≥50 bp and ≥90% identity in 3'UTRs across diverse mammals. CFs are neither composed of simple repeats nor low-complexity regions common to mammalian genomes. Using a transformer-based model, CFs are characterized as A/T-rich and distinguishable from the 3'UTR background. CFs of 100 genes possess conserved RNA structures or are depleted of variation or both. Intriguingly, these genes are enriched in neuronal tissues and play roles in neurodevelopment and RNA processing. Our findings expand on existing studies that attribute enhancer function to UCEs, suggesting a new avenue for exploring the biological roles of CFs in 3'UTRs.