Wfold: A new method for predicting RNA secondary structure with deep learning.

Precise estimations of RNA secondary structures have the potential to reveal the various roles that non-coding RNAs play in regulating cellular activity. However, the mainstay of traditional RNA secondary structure prediction methods relies on thermos-dynamic models via free energy minimization, a laborious process that requires a lot of prior knowledge. Here, RNA secondary structure prediction using Wfold, an end-to-end deep learning-based approach, is suggested. Wfold is trained directly on annotated data and base-pairing criteria. It makes use of an image-like representation of RNA sequences, which an enhanced U-net incorporated with a transformer encoder can process effectively. Wfold eventually increases the accuracy of RNA secondary structure prediction by combining the benefits of self-attention mechanism's mining of long-range information with U-net's ability to gather local information. We compare Wfold's performance using RNA datasets that are within and across families. When trained and evaluated on different RNA families, it achieves a similar performance as the traditional methods, but dramatically outperforms the state-of-the-art methods on within-family datasets. Moreover, Wfold can also reliably forecast pseudoknots. The findings imply that Wfold may be useful for improving sequence alignment, functional annotations, and RNA structure modeling.

Chikungunya Virus RNA Secondary Structures Impact Defective Viral Genome Production.

Chikungunya virus (CHIKV) is a mosquito-borne RNA virus that poses an emerging threat to humans. In a manner similar to other RNA viruses, CHIKV encodes an error-prone RNA polymerase which, in addition to producing full-length genomes, gives rise to truncated, non-functional genomes, which have been coined defective viral genomes (DVGs). DVGs have been intensively studied in the context of therapy, as they can inhibit viral replication and dissemination in their hosts. In this work, we interrogate the influence of viral RNA secondary structures on the production of CHIKV DVGs. We experimentally map RNA secondary structures of the CHIKV genome using selective 2'-hydroxyl acylation analyzed by primer extension and mutational profiling (SHAPE-MaP), which couples chemical labelling with next-generation sequencing. We correlate the inferred secondary structure with preferred deletion sites of CHIKV DVGs. We document an increased probability of DVG generation with truncations at unpaired nucleotides within the secondary structure. We then generated a CHIKV mutant bearing synonymous changes at the nucleotide level to disrupt the existing RNA secondary structure (CHIKV-D2S). We show that CHIKV-D2S presents altered DVG generation compared to wild-type virus, correlating with the change in RNA secondary structure obtained by SHAPE-MaP. Our work thus demonstrates that RNA secondary structure impacts CHIKV DVG production during replication.

The separation between mRNA-ends is more variable than expected.

Effective circularization of mRNA molecules is a key step for the efficient initiation of translation. Research has shown that the intrinsic separation of the ends of mRNA molecules is rather small, suggesting that intramolecular arrangements could provide this effective circularization. Considering that the innate proximity of RNA ends might have important unknown biological implications, we aimed to determine whether the close proximity of the ends of mRNA molecules is a conserved feature across organisms and gain further insights into the functional effects of the proximity of RNA ends. To do so, we studied the secondary structure of 274 full native mRNA molecules from 17 different organisms to calculate the contour length (CL) of the external loop as an index of their end-to-end separation. Our computational predictions show bigger variations (from 0.59 to 31.8 nm) than previously reported and also than those observed in random sequences. Our results suggest that separations larger than 18.5 nm are not favored, whereas short separations could be related to phenotypical stability. Overall, our work implies the existence of a biological mechanism responsible for the increase in the observed variability, suggesting that the CL features of the exterior loop could be relevant for the initiation of translation and that a short CL could contribute to the stability of phenotypes.

Alu insertion-mediated dsRNA structure formation with pre-existing Alu elements as a disease-causing mechanism.

We previously identified a homozygous Alu insertion variant (Alu_Ins) in the 3'-untranslated region (3'-UTR) of SPINK1 as the cause of severe infantile isolated exocrine pancreatic insufficiency. Although we established that Alu_Ins leads to the complete loss of SPINK1 mRNA expression, the precise mechanisms remained elusive. Here, we aimed to elucidate these mechanisms through a hypothesis-driven approach. Initially, we speculated that, owing to its particular location, Alu_Ins could independently disrupt mRNA 3' end formation and/or affect other post-transcriptional processes such as nuclear export and translation. However, employing a 3'-UTR luciferase reporter assay, Alu_Ins was found to result in only an ∼50% reduction in luciferase activity compared to wild type, which is insufficient to account for the severe pancreatic deficiency in the Alu_Ins homozygote. We then postulated that double-stranded RNA (dsRNA) structures formed between Alu elements, an upstream mechanism regulating gene expression, might be responsible. Using RepeatMasker, we identified two Alu elements within SPINK1's third intron, both oriented oppositely to Alu_Ins. Through RNAfold predictions and full-length gene expression assays, we investigated orientation-dependent interactions between these Alu repeats. We provide compelling evidence to link the detrimental effect of Alu_Ins to extensive dsRNA structures formed between Alu_Ins and pre-existing intronic Alu sequences, including the restoration of SPINK1 mRNA expression by aligning all three Alu elements in the same orientation. Given the widespread presence of Alu elements in the human genome and the potential for new Alu insertions at almost any locus, our findings have important implications for detecting and interpreting Alu insertions in disease genes.

Evaluating Performance of Different RNA Secondary Structure Prediction Programs Using Self-cleaving Ribozymes.

Accurate identification of the correct, biologically relevant RNA structures is critical to understanding various aspects of RNA biology since proper folding represents the key to the functionality of all types of RNA molecules and plays pivotal roles in many essential biological processes. Thus, a plethora of approaches have been developed to predict, identify, or solve RNA structures based on various computational, molecular, genetic, chemical, or physicochemical strategies. Purely computational approaches hold distinct advantages over all other strategies in terms of the ease of implementation, time, speed, cost, and throughput, but they strongly underperform in terms of accuracy that significantly limits their broader application. Nonetheless, the advantages of these methods led to a steady development of multiple in silico RNA secondary structure prediction approaches including recent deep learning-based programs. Here, we compared the accuracy of predictions of biologically relevant secondary structures of dozens of self-cleaving ribozyme sequences using seven in silico RNA folding prediction tools with tasks of varying complexity. We found that while many programs performed well in relatively simple tasks, their performance varied significantly in more complex RNA folding problems. However, in general, a modern deep learning method outperformed the other programs in the complex tasks in predicting the RNA secondary structures, at least based on the specific class of sequences tested, suggesting that it may represent the future of RNA structure prediction algorithms.

Strategies to improve the design of gapmer antisense oligonucleotide on allele-specific silencing.

Gapmer antisense oligonucleotides (ASOs) hold therapeutic promise for allele-specific silencing, but face challenges in distinguishing between mutant and wild-type transcripts. This study explores new design strategies to enhance ASO specificity, focusing on a common dominant mutation in COL6A3 gene associated with Ullrich congenital muscular dystrophy. Initial gapmer ASO design exhibited high efficiency but poor specificity for the mutant allele. We then adopted a mixmer design, incorporating additional RNA bases based on computational predictions of secondary structures for both mutant and wild-type alleles, aiming to enhance ASO accessibility to mutant transcripts. The mixmer ASO design demonstrated up to a 3-fold increase in specificity compared with the classical gapmer design. Further refinement involved introducing a nucleotide mismatch as a structural modification, resulting in a 10-fold enhancement in specificity compared with the gapmer design and a 3-fold over the mixmer design. Additionally, we identified for the first time a potential role of the RNA-induced silencing complex (RISC), alongside RNase H1, in gapmer-mediated silencing, in contrast with what was observed with mixmer ASOs, where only RNase H1 was involved. In conclusion, this study presents a novel design concept for allele-specific ASOs leveraging mRNA secondary structures and nucleotide mismatching and suggests a potential involvement of RISC in gapmer-mediated silencing.

RNA structure in alternative splicing regulation: from mechanism to therapy.

Alternative splicing is a highly intricate process that plays a crucial role in post-transcriptional regulation and significantly expands the functional proteome of a limited number of coding genes in eukaryotes. Its regulation is multifactorial, with RNA structure exerting a significant impact. Aberrant RNA conformations lead to dysregulation of splicing patterns, which directly affects the manifestation of disease symptoms. In this review, the molecular mechanisms of RNA secondary structure-mediated splicing regulation are summarized, with a focus on the complex interplay between aberrant RNA conformations and disease phenotypes resulted from splicing defects. This study also explores additional factors that reshape structural conformations, enriching our understanding of the mechanistic network underlying structure-mediated splicing regulation. In addition, an emphasis has been placed on the clinical role of targeting aberrant splicing corrections in human diseases. The principal mechanisms of action behind this phenomenon are described, followed by a discussion of prospective development strategies and pertinent challenges.

Comprehensive survey of conserved RNA secondary structures in full-genome alignment of Hepatitis C virus.

Hepatitis C virus (HCV) is a plus-stranded RNA virus that often chronically infects liver hepatocytes and causes liver cirrhosis and cancer. These viruses replicate their genomes employing error-prone replicases. Thereby, they routinely generate a large 'cloud' of RNA genomes (quasispecies) which-by trial and error-comprehensively explore the sequence space available for functional RNA genomes that maintain the ability for efficient replication and immune escape. In this context, it is important to identify which RNA secondary structures in the sequence space of the HCV genome are conserved, likely due to functional requirements. Here, we provide the first genome-wide multiple sequence alignment (MSA) with the prediction of RNA secondary structures throughout all representative full-length HCV genomes. We selected 57 representative genomes by clustering all complete HCV genomes from the BV-BRC database based on k-mer distributions and dimension reduction and adding RefSeq sequences. We include annotations of previously recognized features for easy comparison to other studies. Our results indicate that mainly the core coding region, the C-terminal NS5A region, and the NS5B region contain secondary structure elements that are conserved beyond coding sequence requirements, indicating functionality on the RNA level. In contrast, the genome regions in between contain less highly conserved structures. The results provide a complete description of all conserved RNA secondary structures and make clear that functionally important RNA secondary structures are present in certain HCV genome regions but are largely absent from other regions. Full-genome alignments of all branches of Hepacivirus C are provided in the supplement.

Targeting sgRNA N secondary structure as a way of inhibiting SARS-CoV-2 replication.

SARS-CoV-2 is a betacoronavirus that causes COVID-19, a global pandemic that has resulted in many infections, deaths, and socio-economic challenges. The virus has a large positive-sense, single-stranded RNA genome of ∼30 kb, which produces subgenomic RNAs (sgRNAs) through discontinuous transcription. The most abundant sgRNA is sgRNA N, which encodes the nucleocapsid (N) protein. In this study, we probed the secondary structure of sgRNA N and a shorter model without a 3' UTR in vitro, using the SHAPE (selective 2'-hydroxyl acylation analyzed by a primer extension) method and chemical mapping with dimethyl sulfate and 1-cyclohexyl-(2-morpholinoethyl) carbodiimide metho-p-toluene sulfonate. We revealed the secondary structure of sgRNA N and its shorter variant for the first time and compared them with the genomic RNA N structure. Based on the structural information, we designed gapmers, siRNAs and antisense oligonucleotides (ASOs) to target the N protein coding region of sgRNA N. We also generated eukaryotic expression vectors containing the complete sequence of sgRNA N and used them to screen for new SARS-CoV-2 gene N expression inhibitors. Our study provides novel insights into the structure and function of sgRNA N and potential therapeutic tools against SARS-CoV-2.

Identification and Genomic Characterization of Two Novel Hepatoviruses in Shrews from Yunnan Province, China.

Hepatitis A virus (HAV), a member of the genus Hepatovirus (Picornaviridae HepV), remains a significant viral pathogen, frequently causing enterically transmitted hepatitis worldwide. In this study, we conducted an epidemiological survey of HepVs carried by small terrestrial mammals in the wild in Yunnan Province, China. Utilizing HepV-specific broad-spectrum RT-PCR, next-generation sequencing (NGS), and QNome nanopore sequencing (QNS) techniques, we identified and characterized two novel HepVs provisionally named EpMa-HAV and EpLe-HAV, discovered in the long-tailed mountain shrew (Episoriculus macrurus) and long-tailed brown-toothed shrew (Episoriculus leucops), respectively. Our sequence and phylogenetic analyses of EpMa-HAV and EpLe-HAV indicated that they belong to the species Hepatovirus I (HepV-I) clade II, also known as the Chinese shrew HepV clade. Notably, the codon usage bias pattern of novel shrew HepVs is consistent with that of previously identified Chinese shrew HepV. Furthermore, our structural analysis demonstrated that shrew HepVs differ from other mammalian HepVs in RNA secondary structure and exhibit variances in key protein sites. Overall, the discovery of two novel HepVs in shrews expands the host range of HepV and underscores the existence of genetically diverse animal homologs of human HAV within the genus HepV.

sincFold: end-to-end learning of short- and long-range interactions in RNA secondary structure.

MOTIVATION: Coding and noncoding RNA molecules participate in many important biological processes. Noncoding RNAs fold into well-defined secondary structures to exert their functions. However, the computational prediction of the secondary structure from a raw RNA sequence is a long-standing unsolved problem, which after decades of almost unchanged performance has now re-emerged due to deep learning. Traditional RNA secondary structure prediction algorithms have been mostly based on thermodynamic models and dynamic programming for free energy minimization. More recently deep learning methods have shown competitive performance compared with the classical ones, but there is still a wide margin for improvement. RESULTS: In this work we present sincFold, an end-to-end deep learning approach, that predicts the nucleotides contact matrix using only the RNA sequence as input. The model is based on 1D and 2D residual neural networks that can learn short- and long-range interaction patterns. We show that structures can be accurately predicted with minimal physical assumptions. Extensive experiments were conducted on several benchmark datasets, considering sequence homology and cross-family validation. sincFold was compared with classical methods and recent deep learning models, showing that it can outperform the state-of-the-art methods.

RNA Folding, Mutation, and Detection.

The structure of RNA molecules is absolutely critical to their functions in a biological system. RNA structure is dynamic and changes in response to cellular needs. Within the last few decades, there has been an increased interest in studying the structure of RNA molecules and how they change to support the needs of the cell in different conditions. Selective 2'-hydroxyl acylation-based mutational profiling using high-throughput sequencing is a powerful method to predict the secondary structure of RNA molecules both in vivo and in immunopurified samples. Selective 2'-hydroxyl acylation-based mutational profiling using high-throughput sequencing works by adding bulky groups onto accessible "flexible" bases in an RNA molecule that are not involved in any base-pairing or RNA-protein interactions. When the RNA is reverse transcribed into cDNA, the bulky groups are incorporated as base mutations, which can be compared to an unmodified control to identify the locations of flexible bases. The comparison of sequence data between modified and unmodified samples allows the computer software program (developed to generate reactivity profiles) to generate RNA secondary structure models. These models can be compared in a variety of conditions to determine how specific stimuli influence RNA secondary structures.

Phylogenetic and Chemical Probing Information as Soft Constraints in RNA Secondary Structure Prediction.

Extrinsic, experimental information can be incorporated into thermodynamics-based RNA folding algorithms in the form of pseudo-energies. Evolutionary conservation of RNA secondary structure elements is detectable in alignments of phylogenetically related sequences and provides evidence for the presence of certain base pairs that can also be converted into pseudo-energy contributions. We show that the centroid base pairs computed from a consensus folding model such as RNAalifold result in a substantial improvement of the prediction accuracy for single sequences. Evidence for specific base pairs turns out to be more informative than a position-wise profile for the conservation of the pairing status. A comparison with chemical probing data, furthermore, strongly suggests that phylogenetic base pairing data are more informative than position-specific data on (un)pairedness as obtained from chemical probing experiments. In this context we demonstrate, in addition, that the conversion of signal from probing data into pseudo-energies is possible using thermodynamic structure predictions as a reference instead of known RNA structures.

Comparative RNA Genomics.

Over the last quarter of a century it has become clear that RNA is much more than just a boring intermediate in protein expression. Ancient RNAs still appear in the core information metabolism and comprise a surprisingly large component in bacterial gene regulation. A common theme with these types of mostly small RNAs is their reliance of conserved secondary structures. Large-scale sequencing projects, on the other hand, have profoundly changed our understanding of eukaryotic genomes. Pervasively transcribed, they give rise to a plethora of large and evolutionarily extremely flexible non-coding RNAs that exert a vastly diverse array of molecule functions. In this chapter we provide a-necessarily incomplete-overview of the current state of comparative analysis of non-coding RNAs, emphasizing computational approaches as a means to gain a global picture of the modern RNA world.

LocARNA 2.0: Versatile Simultaneous Alignment and Folding of RNAs.

Generating accurate alignments of non-coding RNA sequences is indispensable in the quest for understanding RNA function. Nevertheless, aligning RNAs remains a challenging computational task. In the twilight-zone of RNA sequences with low sequence similarity, sequence homologies and compatible, favorable (a priori unknown) structures can be inferred only in dependency of each other. Thus, simultaneous alignment and folding (SA&F) remains the gold-standard of comparative RNA analysis, even if this method is computationally highly demanding. This text introduces to the recent release 2.0 of the software package LocARNA, focusing on its practical application. The package enables versatile, fast and accurate analysis of multiple RNAs. For this purpose, it implements SA&F algorithms in a specific, lightweight flavor that makes them routinely applicable in large scale. Its high performance is achieved by combining ensemble-based sparsification of the structure space and banding strategies. Probabilistic banding strongly improves the performance of LocARNA 2.0 even over previous releases, while simplifying its effective use. Enabling flexible application to various use cases, LocARNA provides tools to globally and locally compare, cluster, and multiply aligned RNAs based on optimization and probabilistic variants of SA&F, which optionally integrate prior knowledge, expressible by anchor and structure constraints.

How Parameters Influence SHAPE-Directed Predictions.

The structure of an RNA sequence encodes information about its biological function. Dynamic programming algorithms are often used to predict the conformation of an RNA molecule from its sequence alone, and adding experimental data as auxiliary information improves prediction accuracy. This auxiliary data is typically incorporated into the nearest neighbor thermodynamic model22 by converting the data into pseudoenergies. Here, we look at how much of the space of possible structures auxiliary data allows prediction methods to explore. We find that for a large class of RNA sequences, auxiliary data shifts the predictions significantly. Additionally, we find that predictions are highly sensitive to the parameters which define the auxiliary data pseudoenergies. In fact, the parameter space can typically be partitioned into regions where different structural predictions predominate.

A Guide to Computational Cotranscriptional Folding Featuring the SRP RNA.

Although RNA molecules are synthesized via transcription, little is known about the general impact of cotranscriptional folding in vivo. We present different computational approaches for the simulation of changing structure ensembles during transcription, including interpretations with respect to experimental data from literature. Specifically, we analyze different mutations of the E. coli SRP RNA, which has been studied comparatively well in previous literature, yet the details of which specific metastable structures form as well as when they form are still under debate. Here, we combine thermodynamic and kinetic, deterministic, and stochastic models with automated and visual inspection of those systems to derive the most likely scenario of which substructures form at which point during transcription. The simulations do not only provide explanations for present experimental observations but also suggest previously unnoticed conformations that may be verified through future experimental studies.

Estimating RNA Secondary Structure Folding Free Energy Changes with efn2.

A number of analyses require estimates of the folding free energy changes of specific RNA secondary structures. These predictions are often based on a set of nearest neighbor parameters that models the folding stability of a RNA secondary structure as the sum of folding stabilities of the structural elements that comprise the secondary structure. In the software suite RNAstructure, the free energy change calculation is implemented in the program efn2. The efn2 program estimates the folding free energy change and the experimental uncertainty in the folding free energy change. It can be run through the graphical user interface for RNAstructure, from the command line, or a web server. This chapter provides detailed protocols for using efn2.

Conservation and distribution of the DRACH motif for potential m6A sites in avian leukosis virus subgroup J.

N6-methyladenosine (m6A) methylation is an internal post-transcriptional modification that has been linked to viral multiplication and pathogenicity. To elucidate the conservation patterns of potential 5'-DRACH-3' motifs in avian leukosis virus subgroup J (ALV-J), 149 ALV-J strains (139 isolates from China; ALV-J prototype HPRS-103 from the UK; and 9 strains from the USA, Russia, India, and Pakistan) available in GenBank before December 2023 were retrieved. According to the prediction results of the SRAMP web-server, these ALV-J genomes contained potential DRACH motifs, with the total number ranging from 43 to 64, which were not determined based on the isolation region and time. Conservative analysis suggested that 37 motifs exhibited a conservation of >80%, including 17 motifs with a grading above "high confidence." Although these motifs were distributed in the U5 region of LTRs and major coding regions, they were enriched in the coding regions of p27, p68, p32, and gp85. The most common m6A-motif sequence of the DRACH motif in the ALV-J genome was GGACU. The RNA secondary structure of each conserved motif predicted by SRAMP and RNAstructure web-server was mainly of two types-A-U pair (21/37) and hairpin loop (16/37)-based on the core adenosine. Considering the systematic comparative analysis performed in this study, future thorough biochemical research is warranted to determine the role of m6A modification during the replication and infection of ALV-J. These conservation and distribution analysis of the DRACH motif for potential m6A sites in ALV-J would provide a foundation for the future intervention of ALV-J infection and m6A modification.

Differential Selection for Translation Efficiency Shapes Translation Machineries in Bacterial Species.

Different bacterial species have dramatically different generation times, from 20-30 min in Escherichia coli to about two weeks in Mycobacterium leprae. The translation machinery in a cell needs to synthesize all proteins for a new cell in each generation. The three subprocesses of translation, i.e., initiation, elongation, and termination, are expected to be under stronger selection pressure to optimize in short-generation bacteria (SGB) such as Vibrio natriegens than in the long-generation Mycobacterium leprae. The initiation efficiency depends on the start codon decoded by the initiation tRNA, the optimal Shine-Dalgarno (SD) decoded by the anti-SD (aSD) sequence on small subunit rRNA, and the secondary structure that may embed the initiation signals and prevent them from being decoded. The elongation efficiency depends on the tRNA pool and codon usage. The termination efficiency in bacteria depends mainly on the nature of the stop codon and the nucleotide immediately downstream of the stop codon. By contrasting SGB with long-generation bacteria (LGB), we predict (1) SGB to have more ribosome RNA operons to produce ribosomes, and more tRNA genes for carrying amino acids to ribosomes, (2) SGB to have a higher percentage of genes using AUG as the start codon and UAA as the stop codon than LGB, (3) SGB to exhibit better codon and anticodon adaptation than LGB, and (4) SGB to have a weaker secondary structure near the translation initiation signals than LGB. These differences between SGB and LGB should be more pronounced in highly expressed genes than the rest of the genes. We present empirical evidence in support of these predictions.