Deep Learning in RNA Structure Studies

Abstract

Deep learning, or artificial neural networks, is a type of machine learning algorithm that can decipher underlying relationships from large volumes of data and has been successfully applied to solve structural biology questions, such as RNA structure. RNA can fold into complex RNA structures by forming hydrogen bonds, thereby playing an essential role in biological processes. While experimental effort has enabled resolving RNA structure at the genome-wide scale, deep learning has been more recently introduced for studying RNA structure and its functionality. Here, we discuss successful applications of deep learning to solve RNA problems, including predictions of RNA structures, non-canonical G-quadruplex, RNA-protein interactions and RNA switches. Following these cases, we give a general guide to deep learning for solving RNA structure problems.

Computational Pipeline for Reference-Free Comparative Analysis of RNA 3D Structures Applied to SARS-CoV-2 UTR Models

“…Although RNA prediction and modeling develops rapidly and many new strategies have been introduced so far [ 8 , 9 , 10 , 11 , 12 , 13 , 14 ], the determination of the 3D structure of RNA remains challenging [ 15 ]. Existing RNA tertiary structure prediction methods can produce a number of models for a single sequence.…”

“…They are usually clustered by similarity and filtered to select the best solutions, but the final result is rarely consistent with the native conformation [ 10 , 16 ]. This problem occurs especially for larger molecules (>100 nts), which contain multi-branched loops, non-canonical base-base and base-backbone interactions, and long-range tertiary contacts [ 14 , 17 , 18 , 19 ]. Choosing a good model for designing, testing, confirming, or rejecting chemical and biological hypotheses is complicated due to the deficiency of experimentally solved structures for most RNA molecules and the randomness present in many 3D RNA modeling algorithms [ 18 , 20 ].…”

Cleaving Folded RNA with DNAzyme Agents

“…The problem of substrate accessibility is traditionally solved by searching for the least folded RNA fragments using bioinformatics. However, prediction of secondary, let alone tertiary, RNA structures is a challenging problem [38,39] . So, in practice, the development of Dz agents relies on screening dozens of Dz sequences using short linear substrates, followed by testing the best candidates for suppression of the targeted mRNA in cell culture [19,40–43] .…”

“…However, prediction of secondary, let alone tertiary, RNA structures is a challenging problem. [38,39] So, in practice, the development of Dz agents relies on screening dozens of Dz sequences using short linear substrates, followed by testing the best candidates for suppression of the targeted mRNA in cell culture. [19,[40][41][42][43] Santoro and Joyce in their pioneering investigation stated that the arms/substrate binding energy should be approximately À 8 to À 10 kcal/mol, [16] which corresponds to ~8-10 nt long arms.…”

The role of secondary structures in the functioning of 3′ untranslated regions of mRNA

“…[14] Advances in technology has led to the development of high-throughput sequencing of chemically modified RNA secondary structures in vivo across the genome, [15] while deep learning algorithms trained on experimental data predict unknown RNA structures with higher accuracy. [16] RNA forms a variety of structures, [17] of which the most common 3′-UTR secondary structures are a stem-loop structure (or hairpin) and G-quadruplex. A hairpin consists of a stem with hydrogen-bonded bases, including canonical Watson-Crick base pairs and noncanonical base pairs, [18] a hairpin loop, and can have a bulge varying in size from one to several nucleotides, forming a flexible extrusion.…”