ML-Based RNA Secondary Structure Prediction Methods: A Survey

In recent years, obtaining RNA secondary structure information has played an important role in RNA and gene function research. Although some RNA secondary structures can be gained experimentally, in most cases, efficient, and accurate computational methods are still needed to predict RNA secondary structure. Current RNA secondary structure prediction methods are mainly based on the minimum free energy algorithm, which finds the optimal folding state of RNA in vivo using an iterative method to meet the minimum energy or other constraints. However, due to the complexity of biotic environment, a true RNA structure always keeps the balance of biological potential energy status, rather than the optimal folding status that meets the minimum energy. For short sequence RNA its equilibrium energy status for the RNA folding organism is close to the minimum free energy status; therefore, the minimum free energy algorithm for predicting RNA secondary structure has higher accuracy. Nevertheless, in a longer sequence RNA, constant folding causes its biopotential energy balance to deviate far from the minimum free energy status. This deviation is because of its complex structure and results in a serious decline in the prediction accuracy of its secondary structure. In this paper, we propose a novel RNA secondary structure prediction algorithm using a convolutional neural network model combined with a dynamic programming method to improve the accuracy with large-scale RNA sequence and structure data. We analyze current experimental RNA sequences and structure data to construct a deep convolutional network model, and then we extract implicit features of an effective classification from large-scale data to predict the pairing probability of each base in an RNA sequence. For the obtained probabilities of RNA sequence base pairing, an enhanced dynamic programming method is applied to obtain the optimal RNA secondary structure. Results indicate that our proposed method is superior to the common RNA secondary structure prediction algorithms in predicting three benchmark RNA families. Based on the characteristics of deep learning algorithm, it can be inferred that the method proposed in this paper has a 30% higher prediction success rate when compared with other algorithms, which will be needed as the amount of real RNA structure data increases in the future.

Nowadays there are numerous discovered natural RNA variations participating in different cellular processes and artificial RNA, e. g., aptamers, riboswitches. One of the required tasks in the investigation of their functions and mechanism of influence on cells and interaction with targets is the prediction of RNA secondary structures. The classic thermodynamic-based prediction algorithms do not consider the specificity of biological folding and deep learning methods that were designed to resolve this issue suffer from homology-based methods problems. Herein, we present a method for RNA secondary structure prediction based on deep learning - AliNA (ALIgned Nucleic Acids). Our method successfully predicts secondary structures for non-homologous to train-data RNA families thanks to usage of the data augmentation techniques. Augmentation extends existing datasets with easily-accessible simulated data. The proposed method shows a high quality of prediction across different benchmarks including pseudoknots. The method is available on GitHub for free (https://github.com/Arty40m/AliNA).

To better elucidate RNA structure-function relationships and to improve the design of pharmaceutical agents that target specific RNA motifs, an understanding of RNA primary, secondary, and tertiary structure is necessary. The prediction of RNA secondary structure from sequence is an intermediate step in predicting RNA three-dimensional structure. RNA secondary structure is typically predicted using a nearest neighbor model based on free energy parameters. The current free energy parameters for 2 × 3 nucleotide loops are based on a 23-member data set of 2 × 3 loops and internal loops of other sizes. A database of representative RNA secondary structures was searched to identify 2 × 3 nucleotide loops that occur in nature. Seventeen of the most frequent 2 × 3 nucleotide loops in this database were studied by optical melting experiments. Fifteen of these loops melted in a two-state manner, and the associated experimental ΔG°(37,2×3) values are, on average, 0.6 and 0.7 kcal/mol different from the values predicted for these internal loops using the predictive models proposed by Lu, Turner, and Mathews [Lu, Z. J., Turner, D. H., and Mathews, D. H. (2006) Nucleic Acids Res. 34, 4912-4924] and Chen and Turner [Chen, G., and Turner, D. H. (2006) Biochemistry 45, 4025-4043], respectively. These new ΔG°(37,2×3) values can be used to update the current algorithms that predict secondary structure from sequence. To improve free energy calculations for duplexes containing 2 × 3 nucleotide loops that still do not have experimentally determined free energy contributions, an updated predictive model was derived. This new model resulted from a linear regression analysis of the data reported here combined with 31 previously studied 2 × 3 nucleotide internal loops. Most of the values for the parameters in this new predictive model are within experimental error of those of the previous models, suggesting that approximations and assumptions associated with the derivation of the previous nearest neighbor parameters were valid. The updated predictive model predicts free energies of 2 × 3 nucleotide internal loops within 0.4 kcal/mol, on average, of the experimental free energy values. Both the experimental values and the updated predictive model can be used to improve secondary structure prediction from sequence.

An Interactive Framework for RNA Secondary Structure Prediction with a Dynamical Treatment of Constraints

Accurate RNA secondary structure information is the cornerstone of gene function research and RNA tertiary structure prediction. However, most traditional RNA secondary structure prediction algorithms are based on the dynamic programming (DP) algorithm, according to the minimum free energy theory, with both hard and soft constraints. The accuracy is particularly dependent on the accuracy of soft constraints (from experimental data like chemical and enzyme detection). With the elongation of the RNA sequence, the time complexity of DP-based algorithms will increase geometrically, as a result, they are not good at coping with relatively long sequences. Furthermore, due to the complexity of the pseudoknots structure, the secondary structure prediction method, based on traditional algorithms, has great defects which cannot predict the secondary structure with pseudoknots well. Therefore, few algorithms have been available for pseudoknots prediction in the past. The ATTfold algorithm proposed in this article is a deep learning algorithm based on an attention mechanism. It analyzes the global information of the RNA sequence via the characteristics of the attention mechanism, focuses on the correlation between paired bases, and solves the problem of long sequence prediction. Moreover, this algorithm also extracts the effective multi-dimensional features from a great number of RNA sequences and structure information, by combining the exclusive hard constraints of RNA secondary structure. Hence, it accurately determines the pairing position of each base, and obtains the real and effective RNA secondary structure, including pseudoknots. Finally, after training the ATTfold algorithm model through tens of thousands of RNA sequences and their real secondary structures, this algorithm was compared with four classic RNA secondary structure prediction algorithms. The results show that our algorithm significantly outperforms others and more accurately showed the secondary structure of RNA. As the data in RNA sequence databases increase, our deep learning-based algorithm will have superior performance. In the future, this kind of algorithm will be more indispensable.

Accurate predictions of RNA secondary structures can help uncover the roles of functional non-coding RNAs. Although machine learning-based models have achieved high performance in terms of prediction accuracy, overfitting is a common risk for such highly parameterized models. Here we show that overfitting can be minimized when RNA folding scores learnt using a deep neural network are integrated together with Turner’s nearest-neighbor free energy parameters. Training the model with thermodynamic regularization ensures that folding scores and the calculated free energy are as close as possible. In computational experiments designed for newly discovered non-coding RNAs, our algorithm (MXfold2) achieves the most robust and accurate predictions of RNA secondary structures without sacrificing computational efficiency compared to several other algorithms. The results suggest that integrating thermodynamic information could help improve the robustness of deep learning-based predictions of RNA secondary structure.

Efforts to use computers in predicting the secondary structure of proteins based only on primary structure information started over a quarter century ago [1-3]. Although the results were encouraging initially, the accuracy of the pioneering methods generally did not attain the level required for using predictions of secondary structures reliably in modelling the three-dimensional topology of proteins. During the last decade, however, the introduction of new computational techniques as well as the use of multiple sequence information has lead to a dramatic increase in the success rate of prediction methods, such that successful 3D modelling based on predicted secondary structure has become feasible [e.g., Ref 4]. This review is aimed at presenting an overview of the scale of the secondary structure prediction problem and associated pitfalls, as well as the history of the development of computational prediction methods. As recent successful strategies for secondary structure prediction all rely on multiple sequence information, some methods for accurate protein multiple sequence alignments will also be described. While the main focus is on prediction methods for globular proteins, also the prediction of trans-membrane segments within membrane proteins will be briefly summarised. Finally, an integrated iterative approach tying secondary structure prediction and multiple alignment will be introduced [5].

The determination of RNA secondary structure (RSS) could help understand RNA's functional mechanisms, guiding the design of RNA-based therapeutics, and advancing synthetic biology applications. However, traditional methods such as NMR for determining RSS are typically time-consuming and labor-intensive. As a result, the accurate prediction of RSS remains a fundamental yet unmet need in RNA research. Various deep learning (DL)-based methods achieved improved accuracy over thermodynamic-based methods. However, the over-parameterization nature of DL makes these methods prone to overfitting and thus limits their generalizability. Meanwhile, the inconsistency of RSS predictions between these methods further aggravated the crisis of generalizability. Here, we propose TrioFold to achieve enhanced generalizability of RSS prediction by integrating base-pairing clues learned from both thermodynamic- and DL-based methods by ensemble learning and convolutional block attention mechanism. TrioFold achieves higher accuracy in intra-family predictions and enhanced generalizability in inter-family and cross-RNA-types predictions. Additionally, we have developed an online webserver equipped with widely used RSS prediction algorithms and analysis tools, providing an accessible platform for the RNA research community. This study demonstrated new opportunities to improve generalizability for RSS predictions by efficient ensemble learning of base-pairing clues learned from both thermodynamic- and DL-based algorithms.

Stochastic context-free grammars (SCFGs) have been applied to predicting RNA secondary structure. The prediction of RNA secondary structure can be facilitated by incorporating with comparative sequence analysis. However, most of existing SCFG-based methods lack explicit phylogenic analysis of homologous RNA sequences, which is probably the reason why these methods are not ideal in practical application. Hence, we present a new SCFG-based method by integrating phylogenic analysis with the newly defined profile SCFG. The method can be summarized as: 1) we define a new profile SCFG, M, to depict consensus secondary structure of multiple RNA sequence alignment; 2) we introduce two distinct hidden Markov models, λ and λ′, to perform phylogenic analysis of homologous RNA sequences. Here, λ is for non-structural regions of the sequence and λ′, is for structural regions of the sequence; 3) we merge λ and λ′, into M to devise a combined model for prediction of RNA secondary structure. We tested our method on data sets constructed from the Rfam database. The sensitivity and specificity of our method are more accurate than those of the predictions by Pfold.

Recent years, RNA secondary structure prediction has attracted much attention of many researchers, which is an important way to grasp the biochemical function of RNA. However, it is very difficult to predict the RNA secondary structure including pseudoknot, which has been identified to be an NP-complete problem. In this paper, a novel multimodal optimization evolutionary algorithm is proposed to optimize the decision space based on the minimum free energy to predict the secondary structure of RNA. Because there exist multiple equivalent secondary structures which represent the same minimum free energy, our algorithm maintain diversity in decision space to find multiple sets of secondary structure simultaneously. The performance of our algorithm is evaluated by PseudoBase instances and compared with some good prediction algorithms. The comparison results show that our method has higher accuracy in RNA secondary structure prediction.

The structure of RNA is very important in biological processes. Over the recent years, lots of machine learning method have been emerged to predict the secondary structure of RNA. In this paper, we use Support Vector Machine to predict secondary structure of RNA sequence. Meanwhile, a sequence-based method is proposed by combining a new feature representation which is based on RNA long-range interaction. We first quote E-NSSEL labels to represent the secondary structure of RNA. Combining with the definition of a new feature vector based on long-range interaction, the secondary structure of test sequence is predicted by SVM model, and the corresponding E-NSSEL sequence is consequently obtained. This sequence can be restored to secondary structure finally. The results which are obtained from RNA training and testing datasets show that this long-range-sequence-based method is superior to those method without new feature. It has higher prediction accuracy as considering the new feature. Moreover, it can predict RNA sequences with long length, which is difficult to deal with traditional folding prediction. Furthermore, it suggests that our method may provide a reliable tool for RNA secondary structure prediction, including the prediction of RNA with pseudoknots.

RNA Secondary Structure Prediction (RSSP) is an optimization problem, where a stable secondary structure is acquired from an RNA primary sequence. Many exact, heuristic and metaheuristic algorithms established in recent years to solve the RSSP problem. We have resolved an accession based on metaheuristic algorithm named Fruit Fly Optimization algorithm to solve the RNA secondary structure prediction problem. FOA is a population-based metaheuristic that works better than all other related existing algorithms and has been employed in different optimization problems. We have redesigned the operators of the FOA algorithm and calculated the minimum Gibbs free energy (ΔG) of the structure to solve the RNA secondary structure problem. We have a Repair function which is known as novel operator that is used to verify and expel the repeated stem from RNA sequence, which is very time-efficient. Every quality of the solutions and spending time are calculated in designing the operators and the repair function. The raised methodology gives efficiency, robustness, and effectiveness in solving the RSSP problem.

Prediction of RNA secondary structure by free energy minimization

RNA secondary structures are known to be important in many biological processes. Many available programs have been developed for RNA secondary structure prediction. Based on our knowledge, however, there still exist secondary structures of known RNA sequences which cannot be covered by these algorithms. In this paper, we provide an efficient algorithm that can handle all RNA secondary structures found in Rfam database. We designed a new stochastic context-free grammar named Rectangle Tree Grammar (RTG) which significantly expands the classes of structures that can be modelled. Our algorithm runs in O(n 6) time and the accuracy is reasonably high, with average PPV and sensitivity over 75%. In addition, the structures that RTG predicts are very similar to the real ones.

RNA structure is hierarchical. Secondary structure contacts, i.e. the canonical base pair contacts, are generally stronger and form faster than the tertiary structure. Therefore, RNA secondary structures can be predicted independently of tertiary structure prediction. Furthermore, the stability of a given RNA secondary structure can be quantified using nearest neighbor free energy parameters. These parameters are the basis of a number of free energy minimization algorithms that predict RNA secondary structure for either a single sequence or multiple sequences. This article reviews the progress of RNA secondary structure prediction by free energy minimization and describes many of the algorithms that have been developed.