Computational design of circular RNA vaccines for enhanced stability
Introduction
Circular RNA (circRNA) molecules have emerged as a distinct class of RNA vaccines owing to their unique covalent closed-loop topology, which confers exceptional thermodynamic stability and resistance to exonuclease degradation compared to linear mRNA [1]. In veterinary vaccinology, circRNA platforms offer particular advantages for thermostable vaccine deployment in field conditions where cold chain continuity is often compromised. The computational design of circRNA vaccines requires integration of bioinformatics pipelines, artificial intelligence (AI) optimization, and biophysical modeling to maximize translation efficiency and immunogenicity [2, 3]. This article provides a detailed technical examination of the methods and algorithms used to engineer circRNA vaccines with enhanced stability, with emphasis on circularization chemistries, secondary structure prediction, and visualization of base-pairing networks.
Circularization methods and topological design
The generation of functional circRNA vaccines depends on efficient and precise circularization of linear precursor RNA. Two principal approaches are employed: enzymatic circularization using group I intron self-splicing (permuted intron-exon system, PIE) and chemical ligation strategies [1]. The PIE method exploits the autocatalytic activity of ribozymes derived from group I introns (e.g., from Tetrahymena thermophila). In this system, the coding sequence is flanked by half-intron sequences; upon splicing, the 3' end is covalently joined to the 5' end, producing a closed circle. The efficiency of this process is highly dependent on the sequence context and secondary structure of the flanking intronic elements.
Chemical circularization methods, such as cyanogen bromide-mediated or enzymatic ligation using T4 RNA ligase, offer alternative routes but often suffer from lower yields or require protective capping strategies [1]. Recent advances in chemical topology include the incorporation of a multicapped design, where additional synthetic cap analogues are positioned at internal sites of the circRNA to enhance translation initiation without disrupting circular integrity [1]. This chemical and topological engineering directly impacts the vaccine's translational output and stability in cellular environments.
Exonuclease resistance and translational augmentation
The primary biophysical advantage of circRNA vaccines lies in their inherent resistance to 5' to 3' exonucleases, which rapidly degrade linear RNA species. The closed loop structure eliminates free ends, thereby circumventing the primary pathway of RNA decay [1]. Additionally, circRNA molecules can be engineered to include specific internal ribosome entry sites (IRES) or modified cap analogues that recruit translation initiation complexes more efficiently. The chemical design of multicapped circRNA has been shown to augment translation levels by several orders of magnitude relative to uncapped circular RNA, without compromising the closed topology [1]. Computational models predict that optimal spacing of IRES elements and cap analogues along the circular sequence reduces steric hindrance and promotes ribosome recycling.
Modeling circular RNA secondary structure
Predicting the secondary structure of circRNA is more computationally demanding than for linear RNA due to the cyclization constraint. Standard free energy minimization algorithms (e.g., based on the nearest-neighbor thermodynamic model) must be adapted to enforce base pairing between the 5' and 3' ends, which become covalently linked [3]. Bioinformatics pipelines for circRNA vaccine design typically incorporate tools that fold the linear precursor under the constraint that terminal nucleotides are adjacent. The resulting minimum free energy (MFE) structure is then used to identify potential IRES accessibility, ribosomal landing pads, and regions of stable duplex that may impede translation.
For large coding sequences (e.g., viral glycoprotein antigens of interest in veterinary pathogens such as Highly Pathogenic Avian Influenza (H5N1) in Poultry and Wild Birds), the secondary structure must be partitioned into domains. Machine learning models trained on experimentally validated circRNA structures can predict context-dependent folding patterns [3]. A key parameter is the circularization efficiency score, which estimates how readily the precursor RNA will adopt a conformation conducive to splicing [2]. This score integrates sequence features such as GC content, local RNA duplex stability, and the presence of structural motifs that inhibit ribozyme activity.
AI-driven sequence optimization
Artificial intelligence has been applied to multiple facets of circRNA vaccine design, including sequence optimization for enhanced stability, codon usage adaptation to the target host species, and selection of optimal IRES and untranslated region (UTR) elements [2]. Multimodal collaborative optimization frameworks combine deep learning predictors of RNA stability and translation with generative models that propose novel sequence variants [2]. In a veterinary context, these models can be trained on host-specific datasets (e.g., chicken, swine, bovine transcriptomes) to maximize antigen expression in target species without altering the encoded protein sequence.
Codon optimization for circRNA differs from linear mRNA because the circular form does not have a poly(A) tail and relies exclusively on IRES-dependent or cap-dependent initiation. AI models have been developed to predict the efficiency of IRES elements in a wide range of species, including livestock and poultry [3]. These models also account for secondary structure at the IRES-flanking junctions, which critically influences ribosome recruitment.
Visualizing RNA loop structures in a 3D Protein Viewer
The three-dimensional conformation of circRNA vaccines can be predicted using coarse-grained or all-atom molecular dynamics simulations. Integration with a 3D Protein Viewer enables researchers to inspect base-pairing networks, loop geometries, and steric interactions between the RNA and translational machinery. For a given circRNA sequence, the following steps are performed:
- Secondary structure prediction using constrained MFE algorithms [3].
- Conversion of the secondary structure to three-dimensional coordinates using RNA structure prediction servers (e.g., based on fragment assembly of known RNA motifs).
- Visualization of the resulting PDB file in the viewer, highlighting the continuous phosphate backbone, base stacking, and hydrogen bonds.
Users can rotate, zoom, and color-code regions such as IRES elements, coding sequences, and chemically modified caps. This visual analysis aids in identifying structural conflicts that may reduce translation efficiency or immunogenicity [2, 1].
Workflow for computational circRNA vaccine design
The integrated computational pipeline is summarized in Figure 1.
flowchart TD
A[Target antigen selection], > B[Sequence retrieval and codon optimization]
B, > C[IRES and UTR selection]
C, > D[Circularization method choice: PIE or chemical]
D, > E[Secondary structure prediction with end constraint]
E, > F[AI-driven optimization of stability and translation]
F, > G[Circularization efficiency scoring]
G, > H[3D structural modeling and visualization]
H, > I[In silico translation efficiency prediction]
I, > J[Experimental validation in target host cells]
Figure 1. Computational workflow for engineering circRNA vaccines with enhanced stability. Steps incorporate AI optimization [2], bioinformatics structure prediction [3], and chemical topology design [1].
Veterinary applications and pathogen targets
CircRNA vaccines are being explored for several veterinary pathogens where rapid antigenic variation or thermostability requirements are critical. Notable examples include:
- Highly Pathogenic Avian Influenza (H5N1): The hemagglutinin (HA) gene can be encoded in a circular transcript to enable stable expression in poultry cells.
- Infectious Coryza in Poultry and Ducks: Subunit antigens from Avibacterium paragallinarum can be presented on circRNA platforms to induce mucosal immunity.
- Porcine Reproductive and Respiratory Syndrome: GP5 and M protein sequences have been evaluated in circRNA formats to improve vaccine shelf life in swine operations.
In each case, the computational design must account for host-specific codon usage, RNA secondary structure preferences, and the optimal IRES for the target species. AI models that are trained on cross-species translation data provide a systematic approach to these adjustments [2, 3].
Conclusion
Computational design of circular RNA vaccines represents a convergence of RNA biochemistry, structural bioinformatics, and artificial intelligence. The ability to engineer enhanced stability through covalent circularization, chemical capping, and sequence optimization has direct implications for veterinary vaccinology, particularly in settings where cold chain reliability is limited. Future developments will likely focus on species-specific IRES libraries and machine learning models that can simultaneously optimize multiple biophysical parameters for a given antigen and host.
References
[1] Chen H, Liu D, Aditham A, et al. Chemical and topological design of multicapped mRNA and capped circular RNA to augment translation. Nat Biotechnol. 2025. URL: https://pubmed.ncbi.nlm.nih.gov/39313647/ *** Disclaimer: This article is for educational and informational purposes only. It is not intended to substitute for professional veterinary advice, diagnosis, treatment, or regulatory guidance. Always consult a licensed veterinarian or qualified specialist regarding animal health, disease diagnosis, and therapeutic decisions.
[2] Zhao Y, Wang H. Artificial intelligence-driven circRNA vaccine development: multimodal collaborative optimization and a new paradigm for biomedical applications. Brief Bioinform. 2025. URL: https://pubmed.ncbi.nlm.nih.gov/40483546/
[3] Liu X, Wang S, Sun Y, et al. Unlocking the potential of circular RNA vaccines: a bioinformatics and computational biology perspective. EBioMedicine. 2025. URL: https://pubmed.ncbi.nlm.nih.gov/40112741/