RNA-MobiSeq: Deep mutational scanning and mobility-based selection for RNA structure inference

Abstract

Tertiary structures of RNAs are increasingly found to be essential for their functions and yet are notoriously challenging to be determined by experimental techniques or predicted by energy-based and AlphaFold2-like deep-learning-based computational methods. Here, we coupled deep mutational scanning with mobility-based selection to separate structurally stable from nonstable RNA variants. The subsequent high-throughput sequencing allows the detection of the covariational signals of key secondary and tertiary base pairs with significantly improved, RNA-specific, unsupervised analysis of covariation-induced deviation of activity (CODA2) and amplification of these signals by Monte-Carlo simulated annealing. The resulting base-pairing structures can serve as the restraints for final energy-based structure predictions. This MobiSeq technique was tested on four structurally different RNAs. The results show that the detection of tertiary base pairs by MobiSeq allows reasonably accurate prediction of RNA tertiary structures. The proposed method should be useful for structural inference and improved mechanistic understanding of all RNAs with detectable changes in mobility due to deleterious mutations.