Computational investigations of structure probing experiments for RNA structure prediction

Ribonucleic acids (RNA) transcripts, and in particular non-coding RNAs, play fundamental roles in cellular metabolism, as they are involved in protein synthesis, catalysis, and regulation of gene expression. In some cases, an RNA’s biological function is mostly dependent on a specific active conformation, making the identification of this single stable structure crucial to identify the role of the RNA and the relationships between its mutations and diseases. On the other hand, RNAs are often found in a dynamic equilibrium of multiple interconverting conformations, that is necessary to regulate their functional activity. In these cases it becomes fundamental to gain knowledge of RNA’s structural ensembles, in order to fully determine its mechanism of action. The current structure determination techniques, both for single-state models such as X-ray crystallography, and for multi-state models such as nuclear magnetic resonance and single-molecule methods, despite proving accurate and reliable in many cases, are extremely slow and costly. In contrast, chemical probing is a class of experimental techniques that provide structural information at single-nucleotide resolution at significantly lower costs in terms of time and required infrastructures. In particular, selective 2′ hydroxyl acylation analyzed via primer extension (SHAPE) has proved a valid chemical mapping technique to probe RNA structure even in vivo. This thesis reports a systematic investi- gation of chemical probing experiments based on two different approaches. The first approach, presented in Chapter 2, relies on machine-learning techniques to optimize a model for mapping experimental data into structural information. The model relies also on co-evolutionary data, in the form of direct coupling analysis (DCA) couplings. The inclusion of this kind of data is chosen in the same spirit of reducing the costs of structure probing, as co-evolutionary analysis relies only on sequencing techniques. The resulting model is proposed as a candidate standard tool for prediction of RNA secondary structure, and some insight in the mechanism of chemical probing is gained by interpreting back its features. Importantly, this work has been developed in the per- spective of building a framework for future refinement and improvement. In this spirit, all the used data and scripts are available at https://github.com/bussilab/shape-dca-data, and the model can be easily retrained and adapted to incorporate arbitrary experimental informa- tion. As the interpretation of the model features suggests the possible emergence of cooperative effects involving RNA nucleotides interacting with SHAPE reagents, a second approach based on Molecular Dynamics simulations is proposed to investigate this hypothesis. The results, along with an originally developed methodology to analyse Molecular Dynamics simulations at variable number of particles, are presented in Chapter 3.