Modeling and automation of sequencing-based characterization of RNA structure

Abstract

Sequence census methods reduce molecular measurements such as transcript abundance and protein-nucleic acid interactions to counting problems via DNA sequencing. We focus on a novel assay utilizing this approach, called selective 2'-hydroxyl acylation analyzed by primer extension sequencing (SHAPE-Seq), that can be used to characterize RNA secondary and tertiary structure. We describe a fully automated data analysis pipeline for SHAPE-Seq analysis that includes read processing, mapping, and structural inference based on a model of the experiment. Our methods rely on the solution of a series of convex optimization problems for which we develop efficient and effective numerical algorithms. Our results can be easily extended to other chemical probes of RNA structure, and also generalized to modeling polymerase drop-off in other sequence census-based experiments.

Fig. 1.

Overview of SHAPE-Seq. As the reverse transcriptase (blue oval) transcribes the RNA, it encounters the first adduct and drops off (Left), or may drop off prematurely (Right). Sequencing of fragments produces data in the form of fragment counts (X). Similarly, a control experiment (Right) measures natural drop-off (fragments labeled Y). The model parameters consist of the adduct probabilities (Θ), the Poisson rate for the number of adducts per molecule (c), and the drop-off probabilities in the control experiment (Γ). Their estimates are denoted by Θ^⋆, Γ^⋆, and c^⋆.

Fig. 2.

ML estimation for S. aureus plasmid pT181 sense RNA. Bootstrap error bars are shown in black on the final ML estimation. Sites 1–45 showed negligible frequencies and were omitted from display. The starred bar is not fully shown and has magnitude -0.16.