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. These methods even yield exact solutions under certain simplifying assumptions. Nevertheless, the accuracy of these classical methods is limited and has seen little improvement over the last decade. This makes it an attractive target for machine learning and consequently several deep learning models have been proposed in recent years. 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 that can not only be generated in arbitrary amounts, but are also guaranteed to be free of biases. 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.

Visualization of RNA secondary structure with pseudoknots

The function of pseudoknots cannot be ignored in the RNA secondary structure. Existing methods for analyzing RNA secondary structures with pseudoknots exhibit many shortcomings. This paper presents a novel RNA secondary structure visualization method in the case of a joint analysis of RNA primary structures and secondary structures. The way is based on the page number representation of the RNA secondary structure. It innovatively uses five vectors to represent bases, which are sequentially connected to outline the characteristics of the RNA secondary structure. The method covers almost all the constituent elements of the RNA secondary structure and extracts features completely. Experiments are based on the available techniques for large-scale annotation of RNA secondary structures, using a combination method of discrete wavelet transform and fractal dimension. The classification effect is compared with the previous RNA secondary structure representation methods. Experimental results show that the RNA secondary structure visualization method proposed in this paper has good application prospects in RNA secondary structure classification.

In Vivo and In Vitro Genome-Wide Profiling of RNA Secondary Structures Reveals Key Regulatory Features in Plasmodium falciparum

It is widely accepted that the structure of RNA plays important roles in a number of biological processes, such as polyadenylation, splicing, and catalytic functions. Dynamic changes in RNA structure are able to regulate the gene expression programme and can be used as a highly specific and subtle mechanism for governing cellular processes. However, the nature of most RNA secondary structures in Plasmodium falciparum has not been determined. To investigate the genome-wide RNA secondary structural features at single-nucleotide resolution in P. falciparum, we applied a novel high-throughput method utilizing the chemical modification of RNA structures to characterize these structures. Structural data from parasites are in close agreement with the known 18S ribosomal RNA secondary structures of P. falciparum and can help to predict the in vivo RNA secondary structure of a total of 3,396 transcripts in the ring-stage and trophozoite-stage developmental cycles. By parallel analysis of RNA structures in vivo and in vitro during the Plasmodium parasite ring-stage and trophozoite-stage intraerythrocytic developmental cycles, we identified some key regulatory features. Recent studies have established that the RNA structure is a ubiquitous and fundamental regulator of gene expression. Our study indicate that there is a critical connection between RNA secondary structure and mRNA abundance during the complex biological programme of P. falciparum. This work presents a useful framework and important results, which may facilitate further research investigating the interactions between RNA secondary structure and the complex biological programme in P. falciparum. The RNA secondary structure characterized in this study has potential applications and important implications regarding the identification of RNA structural elements, which are important for parasite infection and elucidating host-parasite interactions and parasites in the environment.

Graph-Based Analysis of RNA Secondary Structure Similarity Comparison

In organisms, ribonucleic acid (RNA) plays an essential role. Its function is being discovered more and more. Due to the conserved nature of RNA sequences, its function mainly depends on the RNA secondary structure. The discovery of an approximate relationship between two RNA secondary structures helps to understand their functional relationship better. It is an important and urgent task to explore structural similarities from the graphical representation of RNA secondary structures. In this paper, a novel graphical analysis method based on the triple vector curve representation of RNA secondary structures is proposed. A combinational method involving a discrete wavelet transform (DWT) and fractal dimension with sliding window is introduced to analyze and compare the graphs derived from feature extraction; after that, the distance matrix is generated. Then, the distance matrix is analyzed by clustering and visualized as a clustering tree. RNA virus and noncoding RNA datasets are applied to perform experiments and analyze the clustering tree. The results show that the proposed method yields more accurate results in the comparison of RNA secondary structures.

RNA secondary structure prediction using deep learning with thermodynamic integration

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.