Prediction of RNA secondary structure by maximizing pseudo-expected accuracy

LinAliFold and CentroidLinAliFold: fast RNA consensus secondary structure prediction for aligned sequences using beam search methods

Motivation

RNA consensus secondary structure prediction from aligned sequences is a powerful approach for improving the secondary structure prediction accuracy. However, because the computational complexities of conventional prediction tools scale with the cube of the alignment lengths, their application to long RNA sequences, such as viral RNAs or long non-coding RNAs, requires significant computational time.

Results

In this study, we developed LinAliFold and CentroidLinAliFold, fast RNA consensus secondary structure prediction tools based on minimum free energy and maximum expected accuracy principles, respectively. We achieved software acceleration using beam search methods that were successfully used for fast secondary structure prediction from a single RNA sequence. Benchmark analyses showed that LinAliFold and CentroidLinAliFold were much faster than the existing methods while preserving the prediction accuracy. As an empirical application, we predicted the consensus secondary structure of coronaviruses with approximately 30 000 nt in 5 and 79 min by LinAliFold and CentroidLinAliFold, respectively. We confirmed that the predicted consensus secondary structure of coronaviruses was consistent with the experimental results.

Availability and implementation

The source codes of LinAliFold and CentroidLinAliFold are freely available at https://github.com/fukunagatsu/LinAliFold-CentroidLinAliFold.

Supplementary information

Supplementary data are available at Bioinformatics Advances online.

Prediction of RNA secondary structure including pseudoknots for long sequences

RNA structural elements called pseudoknots are involved in various biological phenomena including ribosomal frameshifts. Because it is infeasible to construct an efficiently computable secondary structure model including pseudoknots, secondary structure prediction methods considering pseudoknots are not yet widely available. We developed IPknot, which uses heuristics to speed up computations, but it has remained difficult to apply it to long sequences, such as messenger RNA and viral RNA, because it requires cubic computational time with respect to sequence length and has threshold parameters that need to be manually adjusted. Here, we propose an improvement of IPknot that enables calculation in linear time by employing the LinearPartition model and automatically selects the optimal threshold parameters based on the pseudo-expected accuracy. In addition, IPknot showed favorable prediction accuracy across a wide range of conditions in our exhaustive benchmarking, not only for single sequences but also for multiple alignments.

RNA Secondary Structures with Limited Base Pair Span: Exact Backtracking and an Application

The accuracy of RNA secondary structure prediction decreases with the span of a base pair, i.e., the number of nucleotides that it encloses. The dynamic programming algorithms for RNA folding can be easily specialized in order to consider only base pairs with a limited span L, reducing the memory requirements to O(nL), and further to O(n) by interleaving backtracking. However, the latter is an approximation that precludes the retrieval of the globally optimal structure. So far, the ViennaRNA package therefore does not provide a tool for computing optimal, span-restricted minimum energy structure. Here, we report on an efficient backtracking algorithm that reconstructs the globally optimal structure from the locally optimal fragments that are produced by the interleaved backtracking implemented in RNALfold. An implementation is integrated into the ViennaRNA package. The forward and the backtracking recursions of RNALfold are both easily constrained to structural components with a sufficiently negative z-scores. This provides a convenient method in order to identify hyper-stable structural elements. A screen of the C. elegans genome shows that such features are more abundant in real genomic sequences when compared to a di-nucleotide shuffled background model.

RNA genome conservation and secondary structure in SARS-CoV-2 and SARS-related viruses: a first look

As the COVID-19 outbreak spreads, there is a growing need for a compilation of conserved RNA genome regions in the SARS-CoV-2 virus along with their structural propensities to guide development of antivirals and diagnostics. Here we present a first look at RNA sequence conservation and structural propensities in the SARS-CoV-2 genome. Using sequence alignments spanning a range of betacoronaviruses, we rank genomic regions by RNA sequence conservation, identifying 79 regions of length at least 15 nt as exactly conserved over SARS-related complete genome sequences available near the beginning of the COVID-19 outbreak. We then confirm the conservation of the majority of these genome regions across 739 SARS-CoV-2 sequences subsequently reported from the COVID-19 outbreak, and we present a curated list of 30 "SARS-related-conserved" regions. We find that known RNA structured elements curated as Rfam families and in prior literature are enriched in these conserved genome regions, and we predict additional conserved, stable secondary structures across the viral genome. We provide 106 "SARS-CoV-2-conserved-structured" regions as potential targets for antivirals that bind to structured RNA. We further provide detailed secondary structure models for the extended 5' UTR, frameshifting stimulation element, and 3' UTR. Lastly, we predict regions of the SARS-CoV-2 viral genome that have low propensity for RNA secondary structure and are conserved within SARS-CoV-2 strains. These 59 "SARS-CoV-2-conserved-unstructured" genomic regions may be most easily accessible by hybridization in primer-based diagnostic strategies.

RNA genome conservation and secondary structure in SARS-CoV-2 and SARS-related viruses

As the COVID-19 outbreak spreads, there is a growing need for a compilation of conserved RNA genome regions in the SARS-CoV-2 virus along with their structural propensities to guide development of antivirals and diagnostics. Using sequence alignments spanning a range of betacoronaviruses, we rank genomic regions by RNA sequence conservation, identifying 79 regions of length at least 15 nucleotides as exactly conserved over SARS-related complete genome sequences available near the beginning of the COVID-19 outbreak. We then confirm the conservation of the majority of these genome regions across 739 SARS-CoV-2 sequences reported to date from the current COVID-19 outbreak, and we present a curated list of 30 'SARS-related-conserved' regions. We find that known RNA structured elements curated as Rfam families and in prior literature are enriched in these conserved genome regions, and we predict additional conserved, stable secondary structures across the viral genome. We provide 106 'SARS-CoV-2-conserved-structured' regions as potential targets for antivirals that bind to structured RNA. We further provide detailed secondary structure models for the 5´ UTR, frame-shifting element, and 3´ UTR. Last, we predict regions of the SARS-CoV-2 viral genome have low propensity for RNA secondary structure and are conserved within SARS-CoV-2 strains. These 59 'SARS-CoV-2-conserved-unstructured' genomic regions may be most easily targeted in primer-based diagnostic and oligonucleotide-based therapeutic strategies.

Comparative and integrative analysis of RNA structural profiling data: current practices and emerging questions

BACKGROUND

Structure profiling experiments provide single-nucleotide information on RNA structure. Recent advances in chemistry combined with application of high-throughput sequencing have enabled structure profiling at transcriptome scale and in living cells, creating unprecedented opportunities for RNA biology. Propelled by these experimental advances, massive data with ever-increasing diversity and complexity have been generated, which give rise to new challenges in interpreting and analyzing these data.

RESULTS

We review current practices in analysis of structure profiling data with emphasis on comparative and integrative analysis as well as highlight emerging questions. Comparative analysis has revealed structural patterns across transcriptomes and has become an integral component of recent profiling studies. Additionally, profiling data can be integrated into traditional structure prediction algorithms to improve prediction accuracy.

CONCLUSIONS

To keep pace with experimental developments, methods to facilitate, enhance and refine such analyses are needed. Parallel advances in analysis methodology will complement profiling technologies and help them reach their full potential.

RNA secondary structure prediction from multi-aligned sequences

It has been well accepted that the RNA secondary structures of most functional non-coding RNAs (ncRNAs) are closely related to their functions and are conserved during evolution. Hence, prediction of conserved secondary structures from evolutionarily related sequences is one important task in RNA bioinformatics; the methods are useful not only to further functional analyses of ncRNAs but also to improve the accuracy of secondary structure predictions and to find novel functional RNAs from the genome. In this review, I focus on common secondary structure prediction from a given aligned RNA sequence, in which one secondary structure whose length is equal to that of the input alignment is predicted. I systematically review and classify existing tools and algorithms for the problem, by utilizing the information employed in the tools and by adopting a unified viewpoint based on maximum expected gain (MEG) estimators. I believe that this classification will allow a deeper understanding of each tool and provide users with useful information for selecting tools for common secondary structure predictions.

Locus minimization in breed prediction using artificial neural network approach

Molecular markers, viz. microsatellites and single nucleotide polymorphisms, have revolutionized breed identification through the use of small samples of biological tissue or germplasm, such as blood, carcass samples, embryos, ova and semen, that show no evident phenotype. Classical tools of molecular data analysis for breed identification have limitations, such as the unavailability of referral breed data, causing increased cost of collection each time, compromised computational accuracy and complexity of the methodology used. We report here the successful use of an artificial neural network (ANN) in background to decrease the cost of genotyping by locus minimization. The webserver is freely accessible (http://nabg.iasri.res.in/bisgoat) to the research community. We demonstrate that the machine learning (ANN) approach for breed identification is capable of multifold advantages such as locus minimization, leading to a drastic reduction in cost, and web availability of reference breed data, alleviating the need for repeated genotyping each time one investigates the identity of an unknown breed. To develop this model web implementation based on ANN, we used 51,850 samples of allelic data of microsatellite-marker-based DNA fingerprinting on 25 loci covering 22 registered goat breeds of India for training. Minimizing loci to up to nine loci through the use of a multilayer perceptron model, we achieved 96.63% training accuracy. This server can be an indispensable tool for identification of existing breeds and new synthetic commercial breeds, leading to protection of intellectual property in case of sovereignty and bio-piracy disputes. This server can be widely used as a model for cost reduction by locus minimization for various other flora and fauna in terms of variety, breed and/or line identification, especially in conservation and improvement programs.

Long non-coding RNA and alternative splicing modulations in Parkinson's leukocytes identified by RNA sequencing

The continuously prolonged human lifespan is accompanied by increase in neurodegenerative diseases incidence, calling for the development of inexpensive blood-based diagnostics. Analyzing blood cell transcripts by RNA-Seq is a robust means to identify novel biomarkers that rapidly becomes a commonplace. However, there is lack of tools to discover novel exons, junctions and splicing events and to precisely and sensitively assess differential splicing through RNA-Seq data analysis and across RNA-Seq platforms. Here, we present a new and comprehensive computational workflow for whole-transcriptome RNA-Seq analysis, using an updated version of the software AltAnalyze, to identify both known and novel high-confidence alternative splicing events, and to integrate them with both protein-domains and microRNA binding annotations. We applied the novel workflow on RNA-Seq data from Parkinson's disease (PD) patients' leukocytes pre- and post- Deep Brain Stimulation (DBS) treatment and compared to healthy controls. Disease-mediated changes included decreased usage of alternative promoters and N-termini, 5′-end variations and mutually-exclusive exons. The PD regulated FUS and HNRNP A/B included prion-like domains regulated regions. We also present here a workflow to identify and analyze long non-coding RNAs (lncRNAs) via RNA-Seq data. We identified reduced lncRNA expression and selective PD-induced changes in 13 of over 6,000 detected leukocyte lncRNAs, four of which were inversely altered post-DBS. These included the U1 spliceosomal lncRNA and RP11-462G22.1, each entailing sequence complementarity to numerous microRNAs. Analysis of RNA-Seq from PD and unaffected controls brains revealed over 7,000 brain-expressed lncRNAs, of which 3,495 were co-expressed in the leukocytes including U1, which showed both leukocyte and brain increases. Furthermore, qRT-PCR validations confirmed these co-increases in PD leukocytes and two brain regions, the amygdala and substantia-nigra, compared to controls. This novel workflow allows deep multi-level inspection of RNA-Seq datasets and provides a comprehensive new resource for understanding disease transcriptome modifications in PD and other neurodegenerative diseases.

Long non-coding RNAs (lncRNAs) comprise a novel, fascinating class of RNAs with largely unknown biological functions. Parkinson's-disease (PD) is the most frequent motor disorder, and Deep-brain-stimulation (DBS) treatment alleviates the symptoms, but early disease biomarkers are still unknown and new future genetic interference targets are urgently needed. Using RNA-sequencing technology and a novel computational workflow for in-depth exploration of whole-transcriptome RNA-seq datasets, we detected and analyzed lncRNAs in sequenced libraries from PD patients' leukocytes pre and post-treatment and the brain, adding this full profile resource of over 7,000 lncRNAs to the few human tissues-derived lncRNA datasets that are currently available. Our study includes sample-specific database construction, detecting disease-derived changes in known and novel lncRNAs, exons and junctions and predicting corresponding changes in Polyadenylation choices, protein domains and miRNA binding sites. We report widespread transcript structure variations at the splice junction and exons levels, including novel exons and junctions and alteration of lncRNAs followed by experimental validation in PD leukocytes and two PD brain regions compared with controls. Our results suggest lncRNAs involvement in neurodegenerative diseases, and specifically PD. This comprehensive workflow will be of use to the increasing number of laboratories producing RNA-Seq data in a wide range of biomedical studies.

Fighting against uncertainty: an essential issue in bioinformatics

Many bioinformatics problems, such as sequence alignment, gene prediction, phylogenetic tree estimation and RNA secondary structure prediction, are often affected by the 'uncertainty' of a solution, that is, the probability of the solution is extremely small. This situation arises for estimation problems on high-dimensional discrete spaces in which the number of possible discrete solutions is immense. In the analysis of biological data or the development of prediction algorithms, this uncertainty should be handled carefully and appropriately. In this review, I will explain several methods to combat this uncertainty, presenting a number of examples in bioinformatics. The methods include (i) avoiding point estimation, (ii) maximum expected accuracy (MEA) estimations and (iii) several strategies to design a pipeline involving several prediction methods. I believe that the basic concepts and ideas described in this review will be generally useful for estimation problems in various areas of bioinformatics.

Rise of the RNA machines: exploring the structure of long non-coding RNAs

Novel, profound and unexpected roles of long non-coding RNAs (lncRNAs) are emerging in critical aspects of gene regulation. Thousands of lncRNAs have been recently discovered in a wide range of mammalian systems, related to development, epigenetics, cancer, brain function and hereditary disease. The structural biology of these lncRNAs presents a brave new RNA world, which may contain a diverse zoo of new architectures and mechanisms. While structural studies of lncRNAs are in their infancy, we describe existing structural data for lncRNAs, as well as crystallographic studies of other RNA machines and their implications for lncRNAs. We also discuss the importance of dynamics in RNA machine mechanism. Determining commonalities between lncRNA systems will help elucidate the evolution and mechanistic role of lncRNAs in disease, creating a structural framework necessary to pursue lncRNA-based therapeutics.

CompaRNA: a server for continuous benchmarking of automated methods for RNA secondary structure prediction

We present a continuous benchmarking approach for the assessment of RNA secondary structure prediction methods implemented in the CompaRNA web server. As of 3 October 2012, the performance of 28 single-sequence and 13 comparative methods has been evaluated on RNA sequences/structures released weekly by the Protein Data Bank. We also provide a static benchmark generated on RNA 2D structures derived from the RNAstrand database. Benchmarks on both data sets offer insight into the relative performance of RNA secondary structure prediction methods on RNAs of different size and with respect to different types of structure. According to our tests, on the average, the most accurate predictions obtained by a comparative approach are generated by CentroidAlifold, MXScarna, RNAalifold and TurboFold. On the average, the most accurate predictions obtained by single-sequence analyses are generated by CentroidFold, ContextFold and IPknot. The best comparative methods typically outperform the best single-sequence methods if an alignment of homologous RNA sequences is available. This article presents the results of our benchmarks as of 3 October 2012, whereas the rankings presented online are continuously updated. We will gladly include new prediction methods and new measures of accuracy in the new editions of CompaRNA benchmarks.

A classification of bioinformatics algorithms from the viewpoint of maximizing expected accuracy (MEA)

Many estimation problems in bioinformatics are formulated as point estimation problems in a high-dimensional discrete space. In general, it is difficult to design reliable estimators for this type of problem, because the number of possible solutions is immense, which leads to an extremely low probability for every solution-even for the one with the highest probability. Therefore, maximum score and maximum likelihood estimators do not work well in this situation although they are widely employed in a number of applications. Maximizing expected accuracy (MEA) estimation, in which accuracy measures of the target problem and the entire distribution of solutions are considered, is a more successful approach. In this review, we provide an extensive discussion of algorithms and software based on MEA. We describe how a number of algorithms used in previous studies can be classified from the viewpoint of MEA. We believe that this review will be useful not only for users wishing to utilize software to solve the estimation problems appearing in this article, but also for developers wishing to design algorithms on the basis of MEA.

Analysis of energy-based algorithms for RNA secondary structure prediction

Background

RNA molecules play critical roles in the cells of organisms, including roles in gene regulation, catalysis, and synthesis of proteins. Since RNA function depends in large part on its folded structures, much effort has been invested in developing accurate methods for prediction of RNA secondary structure from the base sequence. Minimum free energy (MFE) predictions are widely used, based on nearest neighbor thermodynamic parameters of Mathews, Turner et al. or those of Andronescu et al. Some recently proposed alternatives that leverage partition function calculations find the structure with maximum expected accuracy (MEA) or pseudo-expected accuracy (pseudo-MEA) methods. Advances in prediction methods are typically benchmarked using sensitivity, positive predictive value and their harmonic mean, namely F-measure, on datasets of known reference structures. Since such benchmarks document progress in improving accuracy of computational prediction methods, it is important to understand how measures of accuracy vary as a function of the reference datasets and whether advances in algorithms or thermodynamic parameters yield statistically significant improvements. Our work advances such understanding for the MFE and (pseudo-)MEA-based methods, with respect to the latest datasets and energy parameters.

Results

We present three main findings. First, using the bootstrap percentile method, we show that the average F-measure accuracy of the MFE and (pseudo-)MEA-based algorithms, as measured on our largest datasets with over 2000 RNAs from diverse families, is a reliable estimate (within a 2% range with high confidence) of the accuracy of a population of RNA molecules represented by this set. However, average accuracy on smaller classes of RNAs such as a class of 89 Group I introns used previously in benchmarking algorithm accuracy is not reliable enough to draw meaningful conclusions about the relative merits of the MFE and MEA-based algorithms. Second, on our large datasets, the algorithm with best overall accuracy is a pseudo MEA-based algorithm of Hamada et al. that uses a generalized centroid estimator of base pairs. However, between MFE and other MEA-based methods, there is no clear winner in the sense that the relative accuracy of the MFE versus MEA-based algorithms changes depending on the underlying energy parameters. Third, of the four parameter sets we considered, the best accuracy for the MFE-, MEA-based, and pseudo-MEA-based methods is 0.686, 0.680, and 0.711, respectively (on a scale from 0 to 1 with 1 meaning perfect structure predictions) and is obtained with a thermodynamic parameter set obtained by Andronescu et al. called BL* (named after the Boltzmann likelihood method by which the parameters were derived).

Conclusions

Large datasets should be used to obtain reliable measures of the accuracy of RNA structure prediction algorithms, and average accuracies on specific classes (such as Group I introns and Transfer RNAs) should be interpreted with caution, considering the relatively small size of currently available datasets for such classes. The accuracy of the MEA-based methods is significantly higher when using the BL* parameter set of Andronescu et al. than when using the parameters of Mathews and Turner, and there is no significant difference between the accuracy of MEA-based methods and MFE when using the BL* parameters. The pseudo-MEA-based method of Hamada et al. with the BL* parameter set significantly outperforms all other MFE and MEA-based algorithms on our large data sets.

IPknot: fast and accurate prediction of RNA secondary structures with pseudoknots using integer programming

MOTIVATION

Pseudoknots found in secondary structures of a number of functional RNAs play various roles in biological processes. Recent methods for predicting RNA secondary structures cover certain classes of pseudoknotted structures, but only a few of them achieve satisfying predictions in terms of both speed and accuracy.

RESULTS

We propose IPknot, a novel computational method for predicting RNA secondary structures with pseudoknots based on maximizing expected accuracy of a predicted structure. IPknot decomposes a pseudoknotted structure into a set of pseudoknot-free substructures and approximates a base-pairing probability distribution that considers pseudoknots, leading to the capability of modeling a wide class of pseudoknots and running quite fast. In addition, we propose a heuristic algorithm for refining base-paring probabilities to improve the prediction accuracy of IPknot. The problem of maximizing expected accuracy is solved by using integer programming with threshold cut. We also extend IPknot so that it can predict the consensus secondary structure with pseudoknots when a multiple sequence alignment is given. IPknot is validated through extensive experiments on various datasets, showing that IPknot achieves better prediction accuracy and faster running time as compared with several competitive prediction methods.

AVAILABILITY

The program of IPknot is available at http://www.ncrna.org/software/ipknot/. IPknot is also available as a web server at http://rna.naist.jp/ipknot/.

CONTACT

satoken@k.u-tokyo.ac.jp; ykato@is.naist.jp

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

Probabilistic alignments with quality scores: an application to short-read mapping toward accurate SNP/indel detection

MOTIVATION

Recent studies have revealed the importance of considering quality scores of reads generated by next-generation sequence (NGS) platforms in various downstream analyses. It is also known that probabilistic alignments based on marginal probabilities (e.g. aligned-column and/or gap probabilities) provide more accurate alignment than conventional maximum score-based alignment. There exists, however, no study about probabilistic alignment that considers quality scores explicitly, although the method is expected to be useful in SNP/indel callers and bisulfite mapping, because accurate estimation of aligned columns or gaps is important in those analyses.

RESULTS

In this study, we propose methods of probabilistic alignment that consider quality scores of (one of) the sequences as well as a usual score matrix. The method is based on posterior decoding techniques in which various marginal probabilities are computed from a probabilistic model of alignments with quality scores, and can arbitrarily trade-off sensitivity and positive predictive value (PPV) of prediction (aligned columns and gaps). The method is directly applicable to read mapping (alignment) toward accurate detection of SNPs and indels. Several computational experiments indicated that probabilistic alignments can estimate aligned columns and gaps accurately, compared with other mapping algorithms e.g. SHRiMP2, Stampy, BWA and Novoalign. The study also suggested that our approach yields favorable precision for SNP/indel calling.

Generalized centroid estimators in bioinformatics

In a number of estimation problems in bioinformatics, accuracy measures of the target problem are usually given, and it is important to design estimators that are suitable to those accuracy measures. However, there is often a discrepancy between an employed estimator and a given accuracy measure of the problem. In this study, we introduce a general class of efficient estimators for estimation problems on high-dimensional binary spaces, which represent many fundamental problems in bioinformatics. Theoretical analysis reveals that the proposed estimators generally fit with commonly-used accuracy measures (e.g. sensitivity, PPV, MCC and F-score) as well as it can be computed efficiently in many cases, and cover a wide range of problems in bioinformatics from the viewpoint of the principle of maximum expected accuracy (MEA). It is also shown that some important algorithms in bioinformatics can be interpreted in a unified manner. Not only the concept presented in this paper gives a useful framework to design MEA-based estimators but also it is highly extendable and sheds new light on many problems in bioinformatics.