RNA Secondary Structure Prediction Research Articles

Assessment of tools for RNA secondary structure prediction and extraction: a final-user perspective

The study of RNA structure is fundamental to clarify the RNA molecular functioning. The flexible RNA nature, the huge number of expressed RNAs, and the variety of functions make it challenging to obtain a quantity of structural information comparable to what is already available for proteins. The in silico prediction of RNA 3D structures is of particular relevance, to understand the fundamental features of the structure-function relationship, because the 3D structure drives the molecular interaction with DNA or protein complexes. The quality of the prediction of the RNA 3D structure is determined by the knowledge of a properly predicted or measured secondary structure. In this paper, we comparatively evaluate computational tools to model RNA secondary structure, focusing our investigation, among the dozens of methods in literature, on tools which are freely available and implemented in web-server versions, providing a more direct access to the final users, not necessarily bioinformatics experts. Our focus is on assessing performances for long sequences, with the final aim of selecting best methods for perspective lncRNAs investigation. Indeed, among RNAs, the non-coding and long non-coding RNAs (lncRNAs, with sequence length larger than 200 nts) assume special relevance, due to their function in regulatory mechanisms, which is still largely unexplored in the case of lncRNAs. As lncRNA experimental structures are at present missing, other families of large RNAs are here used as test cases, to establish the reliability of predictive bioinformatics tools and their perspective applicability to the case of lncRNAs. Communicated by Ramaswamy H. Sarma

BAT-Net: An enhanced RNA Secondary Structure prediction via bidirectional GRU-based network with attention mechanism

Background:RNA Secondary Structure (RSS) has drawn growing concern, both for their pivotal roles in RNA tertiary structures prediction and critical effect in penetrating the mechanism of functional non-coding RNA. Computational techniques that can reduce the in vitro and in vivo experimental costs have become popular in RSS prediction. However, as an NP-hard problem, there is room for improvement that the validity of the prediction RSS with pseudoknots in traditional machine learning predictors. Results:In this essay, by integrating the bidirectional GRU (Gated Recurrent Unit) with the attention, we propose a multilayered neural network called BAT-Net to predict RSS. Different from the state-of-the-art works, BAT-Net can not only make full use of the information about the direct predecessor and direct successor of the predicted base in the RNA sequence but also dynamically adjust the corresponding loss function. The experimental results on five representative datasets extracted from the RNA STRAND database show that the sensitivity, precision, accuracy, and MCC (Matthews Correlation Coefficient) of the BAT-Net have improved by 8.52%, 8.28%, 5.66% and 9.82%, respectively, compared with the benchmark approaches on the best averages. Conclusions:BAT-Net can provide users with more credible RSS results since it has further utilized the source information of the dataset. Comparative results show that the proposed BAT-Net is superior to the other existing methods on the relevant indicators.

RAFFT: Efficient prediction of RNA folding pathways using the fast Fourier transform.

We propose a novel heuristic to predict RNA secondary structure formation pathways that has two components: (i) a folding algorithm and (ii) a kinetic ansatz. This heuristic is inspired by the kinetic partitioning mechanism, by which molecules follow alternative folding pathways to their native structure, some much faster than others. Similarly, our algorithm RAFFT starts by generating an ensemble of concurrent folding pathways ending in multiple metastable structures, which is in contrast with traditional thermodynamic approaches that find single structures with minimal free energies. When we constrained the algorithm to predict only 50 structures per sequence, near-native structures were found for RNA molecules of length ≤ 200 nucleotides. Our heuristic has been tested on the coronavirus frameshifting stimulation element (CFSE): an ensemble of 68 distinct structures allowed us to produce complete folding kinetic trajectories, whereas known methods require evaluating millions of sub-optimal structures to achieve this result. Thanks to the fast Fourier transform on which RAFFT (RNA folding Algorithm wih Fast Fourier Transform) is based, these computations are efficient, with complexity [Formula: see text].

LTPConstraint: a transfer learning based end-to-end method for RNA secondary structure prediction

BackgroundRNA secondary structure is very important for deciphering cell’s activity and disease occurrence. The first method which was used by the academics to predict this structure is biological experiment, But this method is too expensive, causing the promotion to be affected. Then, computing methods emerged, which has good efficiency and low cost. However, the accuracy of computing methods are not satisfactory. Many machine learning methods have also been applied to this area, but the accuracy has not improved significantly. Deep learning has matured and achieves great success in many areas such as computer vision and natural language processing. It uses neural network which is a kind of structure that has good functionality and versatility, but its effect is highly correlated with the quantity and quality of the data. At present, there is no model with high accuracy, low data dependence and high convenience in predicting RNA secondary structure.ResultsThis paper designs a neural network called LTPConstraint to predict RNA secondary structure. The network is based on many network structure such as Bidirectional LSTM, Transformer and generator. It also uses transfer learning to train modelso that the data dependence can be reduced.ConclusionsLTPConstraint has achieved high accuracy in RNA secondary structure prediction. Compared with the previous methods, the accuracy improves obviously both in predicting the structure with pseudoknot and the structure without pseudoknot. At the same time, LTPConstraint is easy to operate and can achieve result very quickly.

MSFF-CDCGAN: A novel method to predict RNA secondary structure based on Generative Adversarial Network.

Access to RNA secondary structure is a prerequisite for understanding and mastering RNA function. RNA secondary structures play an important role in cells, they can cause or contribute to neurological disorders and can be applied in the medical field. However, the experimental method to obtain RNA secondary structure is costly, laborious and not universal. Although computational methods can predict RNA secondary structure more accurately for short-sequence RNAs, it cannot predict long-sequence RNAs and pseudoknot, which is the bottleneck of RNA secondary structure prediction at present. In recent years, researchers have attempted to use deep learning algorithms to predict RNA secondary structure and have achieved results. However, the small amount of data on the secondary structure of long-sequence RNAs leads to the low accuracy of deep learning methods to predict the secondary structure of RNAs across races. Similarly, RNA structure with pseudoknot is very complex and insufficient data caused the deep learning algorithm to struggle to predict the secondary structure of RNA containing pseudoknots. The RNA data are encoded into grayscale images by a unique encoding method based on the real RNA secondary structure and sequence information. Then, this paper reasonably expands the image data to increase the amount of RNA data, solves the problem of insufficient data for predicting long sequences and RNA secondary structure with pseudoknots in current deep learning methods, and provides a good data foundation for deep learning.The article proposes a multi-scale feature fusion Conditional Deep Convolutional Generative Adversarial Network prediction model (MSFF-CDCGAN) based on the improved Conditional Deep Convolutional Generative Adversarial Network (CDCGAN) model to predict RNA secondary structure. The experimental results showed that the MSFF-CDCGAN model could predict long-sequence RNAs and pseudoknots more accurately than traditional prediction methods. This paper introduces Generative Adversarial Network (GAN) to RNA secondary structure prediction for the first time. It uses a unique image encoding approach to expand the original RNA data set, thus transforming the structure prediction problem into an image analysis problem and effectively solving the bottleneck in RNA secondary structure prediction.

Caveats to Deep Learning Approaches to RNA Secondary Structure Prediction.

Machine learning (ML) and in particular deep learning techniques have gained popularity for predicting structures from biopolymer sequences. An interesting case is the prediction of RNA secondary structures, where well established biophysics based methods exist. The accuracy of these classical methods is limited due to lack of experimental parameters and certain simplifying assumptions and has seen little improvement over the last decade. This makes RNA folding an attractive target for machine learning and consequently several deep learning models have been proposed in recent years. However, for ML approaches to be competitive for de-novo structure prediction, the models must not just demonstrate good phenomenological fits, but be able to learn a (complex) biophysical model. In this contribution we discuss limitations of current approaches, in particular due to biases in the training data. Furthermore, we propose to study capabilities and limitations of ML models by first applying them on synthetic data (obtained from a simplified biophysical model) that can be generated in arbitrary amounts and where all biases can be controlled. We assume that a deep learning model that performs well on these synthetic, would also perform well on real data, and vice versa. We apply this idea by testing several ML models of varying complexity. Finally, we show that the best models are capable of capturing many, but not all, properties of RNA secondary structures. Most severely, the number of predicted base pairs scales quadratically with sequence length, even though a secondary structure can only accommodate a linear number of pairs.

Linear-Time Algorithms for RNA Structure Prediction.

RNA secondary structure prediction is widely used to understand RNA function. Existing dynamic programming-based algorithms, both the classical minimum free energy (MFE) methods and partition function methods, suffer from a major limitation: their runtimes scale cubically with the RNA length, and this slowness limits their use in genome-wide applications. Inspired by incremental parsing for context-free grammars in computational linguistics, we designed linear-time heuristic algorithms, LinearFold and LinearPartition, to approximate the MFE structure, partition function and base pairing probabilities. These programs are orders of magnitude faster than Vienna RNAfold and CONTRAfold on long sequences. More interestingly, LinearFold and LinearPartition lead to more accurate predictions on the longest sequence families for which the structures are well established (16S and 23S Ribosomal RNAs), as well as improved accuracies for long-range base pairs (500 + nucleotides apart). This chapter provides protocols for using LinearFold and LinearPartition for secondary structure prediction.

Deep learning models for RNA secondary structure prediction (probably) do not generalize across families.

MotivationThe secondary structure of RNA is of importance to its function. Over the last few years, several papers attempted to use machine learning to improve de novo RNA secondary structure prediction. Many of these papers report impressive results for intra-family predictions but seldom address the much more difficult (and practical) inter-family problem.ResultsWe demonstrate that it is nearly trivial with convolutional neural networks to generate pseudo-free energy changes, modelled after structure mapping data that improve the accuracy of structure prediction for intra-family cases. We propose a more rigorous method for inter-family cross-validation that can be used to assess the performance of learning-based models. Using this method, we further demonstrate that intra-family performance is insufficient proof of generalization despite the widespread assumption in the literature and provide strong evidence that many existing learning-based models have not generalized inter-family.Availability and implementationSource code and data are available at https://github.com/marcellszi/dl-rna.Supplementary information Supplementary data are available at Bioinformatics online.

Secondary structure prediction of long noncoding RNA: review and experimental comparison of existing approaches.

In contrast to messenger RNAs, the function of the wide range of existing long noncoding RNAs (lncRNAs) largely depends on their structure, which determines interactions with partner molecules. Thus, the determination or prediction of the secondary structure of lncRNAs is critical to uncover their function. Classical approaches for predicting RNA secondary structure have been based on dynamic programming and thermodynamic calculations. In the last 4 years, a growing number of machine learning (ML)-based models, including deep learning (DL), have achieved breakthrough performance in structure prediction of biomolecules such as proteins and have outperformed classical methods in short transcripts folding. Nevertheless, the accurate prediction for lncRNA still remains far from being effectively solved. Notably, the myriad of new proposals has not been systematically and experimentally evaluated. In this work, we compare the performance of the classical methods as well as the most recently proposed approaches for secondary structure prediction of RNA sequences using a unified and consistent experimental setup. We use the publicly available structural profiles for 3023 yeast RNA sequences, and a novel benchmark of well-characterized lncRNA structures from different species. Moreover, we propose a novel metric to assess the predictive performance of methods, exclusively based on the chemical probing data commonly used for profiling RNA structures, avoiding any potential bias incorporated by computational predictions when using dot-bracket references. Our results provide a comprehensive comparative assessment of existing methodologies, and a novel and public benchmark resource to aid in the development and comparison of future approaches. Full source code and benchmark datasets are available at: https://github.com/sinc-lab/lncRNA-folding. lbugnon@sinc.unl.edu.ar.

ShapeSorter: a fully probabilistic method for detecting conserved RNAstructure features supported by SHAPE evidence.

There is an increased interest in the determination of RNA structures in vivo as it is now possible to probe them in a high-throughput manner, e.g. using SHAPE protocols. By now, there exist a range of computational methods that integrate experimental SHAPE-probing evidence into computational RNA secondary structure prediction. The state-of-the-art in this field is currently provided by computational methods that employ the minimum-free energy strategy for prediction RNA secondary structures with SHAPE-probing evidence. These methods, however, rely on the assumption that transcripts in vivo fold into the thermodynamically most stable configuration and ignore evolutionary evidence for conserved RNA structure features. We here present a new computational method, ShapeSorter, that predicts RNA structure features without employing the thermodynamic strategy. Instead, ShapeSorter employs a fully probabilistic framework to identify RNA structure features that are supported by evolutionary and SHAPE-probing evidence. Our method can capture RNA structure heterogeneity, pseudo-knotted RNA structures as well as transient and mutually exclusive RNA structure features. Moreover, it estimates P-values for the predicted RNA structure features which allows for easy filtering and ranking. We investigate the merits of our method in a comprehensive performance benchmarking and conclude that ShapeSorter has a significantly superior performance for predicting base-pairs than the existing state-of-the-art methods.

A Test and Refinement of Folding Free Energy Nearest Neighbor Parameters for RNA Including N6-Methyladenosine

RNA folding free energy change parameters are widely used to predict RNA secondary structure and to design RNA sequences. These parameters include terms for the folding free energies of helices and loops. Although the full set of parameters has only been traditionally available for the four common bases and backbone, it is well known that covalent modifications of nucleotides are widespread in natural RNAs. Covalent modifications are also widely used in engineered sequences. We recently derived a full set of nearest neighbor terms for RNA that includes N6-methyladenosine (m6A). In this work, we test the model using 98 optical melting experiments, matching duplexes with or without N6-methylation of A. Most experiments place RRACH, the consensus site of N6-methylation, in a variety of contexts, including helices, bulge loops, internal loops, dangling ends, and terminal mismatches. For matched sets of experiments that include either A or m6A in the same context, we find that the parameters for m6A are as accurate as those for A. Across all experiments, the root mean squared deviation between estimated and experimental free energy changes is 0.67 kcal/mol. We used the new experimental data to refine the set of nearest neighbor parameter terms for m6A. These parameters enable prediction of RNA secondary structures including m6A, which can be used to model how N6-methylation of A affects RNA structure.

RNA folding using quantum computers.

The 3-dimensional fold of an RNA molecule is largely determined by patterns of intramolecular hydrogen bonds between bases. Predicting the base pairing network from the sequence, also referred to as RNA secondary structure prediction or RNA folding, is a nondeterministic polynomial-time (NP)-complete computational problem. The structure of the molecule is strongly predictive of its functions and biochemical properties, and therefore the ability to accurately predict the structure is a crucial tool for biochemists. Many methods have been proposed to efficiently sample possible secondary structure patterns. Classic approaches employ dynamic programming, and recent studies have explored approaches inspired by evolutionary and machine learning algorithms. This work demonstrates leveraging quantum computing hardware to predict the secondary structure of RNA. A Hamiltonian written in the form of a Binary Quadratic Model (BQM) is derived to drive the system toward maximizing the number of consecutive base pairs while jointly maximizing the average length of the stems. A Quantum Annealer (QA) is compared to a Replica Exchange Monte Carlo (REMC) algorithm programmed with the same objective function, with the QA being shown to be highly competitive at rapidly identifying low energy solutions. The method proposed in this study was compared to three algorithms from literature and, despite its simplicity, was found to be competitive on a test set containing known structures with pseudoknots.

Thermodynamic characterization of naturally occurring RNA pentaloops.

RNA folding is hierarchical; therefore, predicting RNA secondary structure from sequence is an intermediate step in predicting tertiary structure. Secondary structure prediction is based on a nearest neighbor model using free energy minimization. To improve secondary structure prediction, all types of naturally occurring secondary structure motifs need to be thermodynamically characterized. However, not all secondary structure motifs are well characterized. Pentaloops, the second most abundant hairpin size, is one such uncharacterized motif. In fact, the current thermodynamic model used to predict the stability of pentaloops was derived from a small data set of pentaloops and from data for other hairpins of different sizes. Here, the most commonly occurring pentaloops were identified and optically melted. New experimental data for 22 pentaloop sequences were combined with previously published data for nine pentaloop sequences. Using linear regression, a pentaloop-specific model was derived. This new model is simpler and more accurate than the current model. The new experimental data and improved model can be incorporated into software that is used to predict RNA secondary structure from sequence.

Rational design of hairpin RNA excited states reveals multi-step transitions

RNA excited states represent a class of high-energy-level and thus low-populated conformational states of RNAs that are sequestered within the free energy landscape until being activated by cellular cues. In recent years, there has been growing interest in structural and functional studies of these transient states, but the rational design of excited states remains unexplored. Here we developed a method to design small hairpin RNAs with predefined excited states that exchange with ground states through base pair reshuffling, and verified these transient states by combining NMR relaxation dispersion technique and imino chemical shift prediction. Using van’t Hoff analysis and accelerated molecular dynamics simulations, a mechanism of multi-step sequential transition has been revealed. The efforts made in this study will expand the scope of RNA rational design, and also contribute towards improved predictions of RNA secondary structure.

RNA secondary structure prediction with convolutional neural networks

BackgroundPredicting the secondary, i.e. base-pairing structure of a folded RNA strand is an important problem in synthetic and computational biology. First-principle algorithmic approaches to this task are challenging because existing models of the folding process are inaccurate, and even if a perfect model existed, finding an optimal solution would be in general NP-complete.ResultsIn this paper, we propose a simple, yet effective data-driven approach. We represent RNA sequences in the form of three-dimensional tensors in which we encode possible relations between all pairs of bases in a given sequence. We then use a convolutional neural network to predict a two-dimensional map which represents the correct pairings between the bases. Our model achieves significant accuracy improvements over existing methods on two standard datasets, RNAStrAlign and ArchiveII, for 10 RNA families, where our experiments show excellent performance of the model across a wide range of sequence lengths. Since our matrix representation and post-processing approaches do not require the structures to be pseudoknot-free, we get similar good performance also for pseudoknotted structures.ConclusionWe show how to use an artificial neural network design to predict the structure for a given RNA sequence with high accuracy only by learning from samples whose native structures have been experimentally characterized, independent of any energy model.

Length-Dependent Deep Learning Model for RNA Secondary Structure Prediction.

Deep learning methods for RNA secondary structure prediction have shown higher performance than traditional methods, but there is still much room to improve. It is known that the lengths of RNAs are very different, as are their secondary structures. However, the current deep learning methods all use length-independent models, so it is difficult for these models to learn very different secondary structures. Here, we propose a length-dependent model that is obtained by further training the length-independent model for different length ranges of RNAs through transfer learning. 2dRNA, a coupled deep learning neural network for RNA secondary structure prediction, is used to do this. Benchmarking shows that the length-dependent model performs better than the usual length-independent model.

Efficient Generation of RNA Secondary Structure Prediction Algorithm Under PAR Framework.

Prediction of RNA secondary structure is an important part of bioinformatics genomics research. Mastering RNA secondary structure can help us to better analyze protein synthesis, cell differentiation, metabolism, and genetic processes and thus reveal the genetic laws of organisms. Comparative sequence analysis, support vector machine, centroid method, and other algorithms in RNA secondary structure prediction algorithms often use dynamic programming algorithm to predict RNA secondary structure because of their huge time and space consumption and complex data structure. In this article, the domain of RNA secondary structure prediction algorithm based on dynamic programming (DP-SSP) is analyzed in depth, and the domain features are modeled. According to the generative programming method, the DP-SSP algorithm components are interactively designed. With the support of PAR platform, the DP-SSP algorithm component library is formally realized. Finally, the concrete algorithm is generated through component assembly, which improves the efficiency and reliability of algorithm development.

The Structural Annotations of The Mir-122 Non-Coding RNA from The Tilapia Fish (Oreochromis niloticus)

Tilapia (Oreochromis niloticus) is an important fisheries commodity. Scientific efforts have been done to increase its quality. One of them is staging a premium diet such as a fat-enriched diet. The transcriptomics approach is able to provide the signatures of the diet outcomes by observing the micro(mi)RNA signature in transcriptional regulation. Hence, it was found that the availability of mir-122 is essential in the regulation of a high-fat diet in tilapia. However, this transcriptomics signature is lacking structural annotations and the complete interaction annotations with its silencing(si)RNA. RNAcentral website was navigated for the latest annotation of mir-122 from tilapia and other species as a comparison. MEGA X was employed to comprehend the miRNA evolutionary repertoire. The RNA secondary structure prediction tools from the Vienna RNA package and the RNA tertiary structure prediction tools from simRNA and modeRNA are secured with default parameters. The HNADOCK tools were leveraged to observe the interaction between mir-122 and its siRNA. The post-processing was conducted with the Chimera visualization tool. The secondary and tertiary structure of the mir-122 and its siRNA could be elucidated, docked, and visualized. In this end, further effort to develop a comprehensive molecular breeding tool could be secured with the structural annotation information.

A reproducibility analysis-based statistical framework for residue-residue evolutionary coupling detection.

Direct coupling analysis (DCA) has been widely used to infer evolutionary coupled residue pairs from the multiple sequence alignment (MSA) of homologous sequences. However, effectively selecting residue pairs with significant evolutionary couplings according to the result of DCA is a non-trivial task. In this study, we developed a general statistical framework for significant evolutionary coupling detection, referred to as irreproducible discovery rate (IDR)-DCA, which is based on reproducibility analysis of the coupling scores obtained from DCA on manually created MSA replicates. IDR-DCA was applied to select residue pairs for contact prediction for monomeric proteins, protein-protein interactions and monomeric RNAs, in which three different versions of DCA were applied. We demonstrated that with the application of IDR-DCA, the residue pairs selected using a universal threshold always yielded stable performance for contact prediction. Comparing with the application of carefully tuned coupling score cutoffs, IDR-DCA always showed better performance. The robustness of IDR-DCA was also supported through the MSA downsampling analysis. We further demonstrated the effectiveness of applying constraints obtained from residue pairs selected by IDR-DCA to assist RNA secondary structure prediction.

SSRTool: A web tool for evaluating RNA secondary structure predictions based on species-specific functional interpretability

RNA secondary structures can carry out essential cellular functions alone or interact with one another to form the hierarchical tertiary structures. Experimental structure identification approa ches can show the in vitro structures of RNA molecules. However, they usually have limits in the resolution and are costly. In silico structure prediction tools are thus primarily relied on for pre-experiment analysis. Various structure prediction models have been developed over the decades. Since these tools are usually used before knowing the actual RNA structures, evaluating and ranking the pile of secondary structure predictions of a given sequence is essential in computational analysis. In this research, we implemented a web service called SSRTool (RNA Secondary Structure prediction Ranking Tool) to assist in the ranking and evaluation of the generated predicted structures of a given sequence. Based on the computed species-specific interpretability significance in four common RNA structure–function aspects, SSRTool provides three functions along with visualization interfaces: (1) Rank user-generated predictions. (2) Provide an automated streamline of structure prediction and ranking for a given sequence. (3) Infer the functional aspects of a given structure. We demonstrated the applicability of SSRTool via real case studies and reported the similar trends between computed species-specific rankings and the corresponding prediction F1 values. The SSRTool web service is available online at https://cobisHSS0.im.nuk.edu.tw/SSRTool/, http://cosbi3.ee.ncku.edu.tw/SSRTool/, or the redirecting site https://github.com/cobisLab/SSRTool/.