Alignment-free discovery of conserved RNA structures in the human genome

Many functional RNAs were first found as conserved RNA structures, and their roles in gene regulation were identified later. In humans and other vertebrates, imperfect genome alignments have hindered the discovery of such structures. Here we present RLMHelix, which selects candidate conserved RNA structures from single sequences before any alignment by scoring helices supported by language-model dependencies and thermodynamics. Through distillation, RLMHelix rapidly scans the transcribed human genome and ranks known RNAs higher than conventional thermodynamic scanning. Two-thirds of 15,821 candidates are new to four earlier genome-wide screens, yet agree with in-cell probing, and over half retain structural signal in homologues. In eCLIP data and in exploratory pull-downs, higher RLMHelix scores were associated with more binding by double-stranded RNA-binding proteins. By linking candidates to their binding proteins, RLMHe-lix yields testable functional hypotheses and illuminates a path from unrecognized human RNA structures to RNA biology and therapeutic targets.