Inferring RNA structure from mobility-based deep mutational landscapes

Abstract Deciphering RNA three-dimensional structures remains a central challenge in molecular biology. While deep mutational profiling captures evolutionary constraints, its application has been limited to a few functions, such as ribozyme self-cleavage, that are compatible with high-throughput selection. Here we introduce RNA-MobiSeq, an integrative platform that leverages electrophoretic mobility as a proxy for native RNA structure. Native gel profiling of deep-mutational libraries enriches for structural homologs, and covariation analysis via an RNA-specific unsupervised method converts these data into effective folding restraints, enabling accurate prediction of base-pairing (F1-score ≥ 90%) and tertiary (3.6–11.2 Å RMSD) structures for nine diverse RNAs, including riboswitches, ribozymes, and long noncoding RNAs. Ranking as the best performer in CASP15-16 RNA targets predicted by the template-free techniques and validated through replicate and oligo-pool experiments, RNA-MobiSeq provides a general, probe-free framework for elucidating RNA structural landscapes, bridging sequence information and functional understanding.