- RNA secondary structure prediction is an important area of bioinformatics that focuses on determining the three-dimensional-independent folding patterns formed by base pairing within RNA molecules. RNA molecules are generally composed of single nucleotide chains, but they can fold into complex structures through interactions between complementary nucleotide bases. These interactions produce characteristic structural elements such as stems, hairpin loops, internal loops, bulges, and multi-branch loops. RNA secondary structure plays an essential role in determining molecular stability, biological activity, molecular recognition, and interactions with proteins and other nucleic acids. Computational prediction of RNA secondary structure helps researchers understand RNA function, identify conserved structural elements, investigate regulatory mechanisms, and analyse noncoding RNAs.
- RNA secondary structure is primarily formed through hydrogen bonding between complementary nucleotides. In conventional Watson–Crick base pairing, adenine (A) pairs with uracil (U), while guanine (G) pairs with cytosine (C). RNA can also form noncanonical interactions, including G–U wobble base pairs, which contribute to RNA folding and structural diversity. These interactions allow RNA molecules to form stems and other structural arrangements. A stem consists of paired nucleotides, while a hairpin loop forms when a single RNA strand folds back on itself and creates a loop at the end of a stem. Internal loops and bulges occur when unpaired nucleotides interrupt otherwise paired regions. Multi-branch loops connect several stem regions and contribute to complex RNA structures.
- RNA secondary structure differs from primary and tertiary structure. Primary structure refers to the linear nucleotide sequence of an RNA molecule. Secondary structure describes local and long-range base-pairing interactions that create stems and loops. Tertiary structure refers to the complete three-dimensional arrangement of the RNA molecule, including interactions between distant regions and structural elements. Secondary structure prediction is therefore an important intermediate step in understanding RNA folding, although it does not provide a complete representation of the molecule’s three-dimensional conformation.
- The biological importance of RNA secondary structure is evident in many RNA classes. Transfer RNA (tRNA) contains a characteristic cloverleaf secondary structure that supports amino acid attachment and recognition during protein synthesis. Ribosomal RNA (rRNA) forms extensive secondary structures that contribute to ribosome organization and function. MicroRNA (miRNA) precursors form characteristic hairpin structures that are processed into mature regulatory RNAs. Small nuclear RNAs (snRNAs) contain structural elements required for spliceosome function, while small nucleolar RNAs (snoRNAs) use conserved structural features to interact with other RNA molecules and proteins. Long noncoding RNAs (lncRNAs) and circular RNAs (circRNAs) may also contain structural regions involved in molecular interactions and gene regulation.
- Computational RNA secondary structure prediction generally uses thermodynamic models to estimate the most favourable folding arrangement of an RNA sequence. The most widely used approach is minimum free energy (MFE) prediction. According to this method, the preferred secondary structure is the one predicted to have the lowest Gibbs free energy under a specified thermodynamic model. RNA folding depends on the stability of base pairs, stacking interactions, loop formation, bulges, and other structural features. Thermodynamic parameters derived from experimental measurements are used to estimate the energy contribution of different structural elements. The predicted MFE structure represents a computational estimate of a favourable folding state rather than a guarantee of the exact structure adopted under all biological conditions.
- The Gibbs free energy of RNA folding can be represented as:
- ΔG = ΔH − TΔS
- Here, ΔG represents the change in Gibbs free energy, ΔH represents the change in enthalpy, T is the absolute temperature, and ΔS represents the change in entropy. A negative ΔG indicates that a folding process is thermodynamically favourable under the specified conditions. RNA folding algorithms use thermodynamic parameters to estimate the contributions of base pairing and structural elements to the overall free energy.
- Dynamic programming is a fundamental computational technique used in RNA secondary structure prediction. It divides the folding problem into smaller subproblems and combines their solutions to identify an optimal structure. Classical algorithms, including the Zuker algorithm, use dynamic programming to calculate minimum free energy structures while considering possible base-pairing arrangements. The computational approach typically uses a simplified representation of secondary structure and incorporates constraints to avoid physically unrealistic interactions. Most standard folding algorithms assume that RNA secondary structures are pseudoknot-free, although specialized methods have been developed to predict pseudoknots.
- Several computational tools are widely used for RNA secondary structure prediction. ViennaRNA is one of the most commonly used software packages and provides tools for RNA folding, energy evaluation, partition function calculations, and RNA structure comparison. RNAfold predicts minimum free energy structures and base-pairing probabilities. RNAstructure provides methods for secondary structure prediction, energy calculations, and analysis of experimental RNA structure data. Mfold is another established tool for predicting RNA secondary structures using thermodynamic models. CONTRAfold uses probabilistic models and machine-learning approaches rather than relying exclusively on conventional thermodynamic parameters. These tools support different types of RNA analysis and may produce different predictions depending on their algorithms, parameters, and assumptions.
- Minimum free energy prediction is not the only approach used in RNA structure analysis. Partition function methods calculate the probability distribution of possible secondary structures rather than identifying only one optimal structure. This approach is valuable because RNA molecules may adopt multiple alternative conformations. Base-pairing probability indicates how frequently a particular nucleotide pair is predicted to occur across the ensemble of possible structures. High-probability base pairs are generally considered more structurally reliable than pairs appearing in only a small proportion of predicted conformations. RNAfold and RNAstructure provide methods for analysing structural ensembles and base-pairing probabilities.
- RNA structure prediction can also incorporate comparative sequence information. When homologous RNA sequences from different species are available, researchers can identify conserved base-pairing patterns and evolutionary relationships. Comparative methods use multiple sequence alignments to detect compensatory mutations, in which changes in paired nucleotides preserve the structural arrangement. Covariance models are particularly useful for this purpose because they combine sequence conservation and secondary structure information. Infernal and Rfam support the identification and classification of RNA families using covariance models. Comparative approaches can often provide stronger evidence for conserved RNA structures than prediction based on a single sequence.
- RNA secondary structure prediction is also important in the analysis of microRNAs and other small RNAs. Primary microRNA transcripts are processed into precursor molecules that form hairpin structures. Computational tools such as RNAfold and RNAstructure can predict these hairpins, while miRDeep2 uses sequencing evidence and precursor structural characteristics to support novel miRNA identification. Similarly, RNA structure analysis can help investigate siRNA precursors, piRNA-related RNA regions, and other regulatory RNA molecules. However, a predicted hairpin structure alone does not establish that a sequence is a functional miRNA or another specific RNA class.
- For long noncoding RNAs, secondary structure prediction helps investigate possible functional regions, molecular interaction sites, and structural conservation. lncRNAs may contain local structural domains that interact with RNA-binding proteins, transcription factors, chromatin-modifying complexes, or other RNA molecules. Because lncRNAs can be long and structurally complex, predicting their complete secondary structures remains computationally challenging. Researchers may therefore focus on shorter functional regions or combine computational predictions with experimental structural information. RNAfold, RNAstructure, and other RNA modelling tools can support these investigations.
- RNA secondary structure prediction is also useful in circular RNA research. Circular RNAs contain covalently closed RNA structures and may form local base-pairing interactions that influence their stability, localization, and molecular interactions. Computational analysis can identify possible structural regions involved in microRNA binding or RNA-binding protein interactions. However, circular RNA folding may require consideration of the circular sequence arrangement, and predictions made using standard linear RNA assumptions may not fully represent its structural properties. Specialized computational approaches and experimental validation can improve the interpretation of circRNA structures.
- Experimental techniques can provide additional evidence for predicted RNA structures. Chemical probing methods such as SHAPE (Selective 2′-Hydroxyl Acylation analyzed by Primer Extension) and DMS (Dimethyl Sulfate) probing provide information about nucleotide accessibility and pairing. Enzymatic probing uses structure-sensitive nucleases to identify accessible and paired RNA regions. SHAPE-MaP and related approaches combine chemical modification with sequencing to investigate RNA structure across large numbers of nucleotides. These experimental data can be incorporated into computational prediction tools to constrain folding models and improve structural accuracy. Other methods, including nuclear magnetic resonance (NMR), X-ray crystallography, and cryo-electron microscopy, can provide detailed structural information for suitable RNA molecules and complexes.
- RNA secondary structure prediction has applications in many areas of molecular biology and bioinformatics. It supports the identification of noncoding RNAs, analysis of RNA family conservation, investigation of gene regulation, and study of RNA–protein interactions. In virology, RNA structural elements can influence viral genome replication, translation, packaging, and host interactions. In medicine, RNA structure analysis contributes to research on genetic disorders, cancer, infectious diseases, and RNA-based therapeutics. RNA structure is also important in the design of antisense oligonucleotides, small interfering RNAs, messenger RNA vaccines, and other RNA-based technologies because molecular folding can influence accessibility, stability, and biological activity.
- In agriculture and animal genomics, RNA secondary structure prediction can help investigate regulatory RNAs associated with growth, development, reproduction, immunity, metabolism, and environmental adaptation. Comparative analysis of RNA structures among livestock species and breeds may reveal conserved regulatory elements and possible functional differences. Structural analysis of miRNAs, lncRNAs, circRNAs, and other noncoding RNAs can be integrated with RNA-seq, small RNA sequencing, quantitative trait loci (QTL), genome-wide association studies (GWAS), and gene-expression networks to investigate the molecular regulation of economically important traits.
- Despite significant advances, RNA secondary structure prediction has several limitations. Most algorithms simplify RNA folding by using thermodynamic assumptions and may not fully account for cellular conditions, molecular crowding, RNA modifications, protein interactions, or interactions with other nucleic acids. Many standard methods do not predict pseudoknots, which are structural arrangements involving crossing base-pair interactions. RNA molecules may also adopt alternative structures depending on temperature, ion concentrations, cellular location, and binding partners. Long RNA molecules create additional computational challenges because the number of possible folding arrangements increases rapidly with sequence length. Consequently, computational predictions should be interpreted as hypotheses and, where possible, combined with experimental structural evidence.
- Overall, RNA secondary structure prediction is an essential component of modern RNA bioinformatics. It enables researchers to investigate RNA folding, identify structural motifs, analyse conserved base-pairing patterns, and predict potential molecular functions. Integration of thermodynamic modelling, dynamic programming, probabilistic methods, comparative genomics, specialized software, and experimental probing provides a powerful framework for understanding RNA structure and function. These approaches continue to advance research in molecular biology, medicine, biotechnology, agriculture, and animal genomics.