Structural Virology and Molecular Dynamics: Predicting Viral Protein Conformations for Antiviral Design

1. Introduction

The application of structural virology to antiviral and vaccine design has become a cornerstone of modern veterinary medicine and emerging zoonotic disease preparedness [1, 2]. Understanding the three-dimensional architecture of viral proteins at atomic or near-atomic resolution is essential for elucidating mechanisms of host cell entry, replication, assembly, and immune evasion [1, 3, 32]. The structural characterization of viral components such as capsid proteins, envelope glycoproteins, replication machinery, and non-structural proteins provides fundamental insights into viral pathogenesis [1, 31, 33]. For veterinary pathogens such as porcine reproductive and respiratory syndrome virus (PRRSV), foot-and-mouth disease virus (FMDV), and various flaviviruses, structural data directly inform the rational design of vaccines and therapeutics [4, 5, 6, 31].

The dynamic nature of viral proteins presents a significant challenge for static structural models [33, 34]. Viral envelope glycoproteins, in particular, undergo substantial conformational changes during receptor binding and membrane fusion [32, 35]. These transitions, from prefusion to postfusion states, are critical targets for neutralizing antibodies and small-molecule inhibitors [3, 35]. Computational approaches, including molecular dynamics (MD) simulations, provide a framework for capturing these dynamic events in silico [29, 35]. By integrating experimental structural data with computational modeling, researchers can predict conformational ensembles, identify cryptic binding sites, and assess the impact of mutations on protein stability and function [7, 8, 30].

This review examines the principal computational methods used for predicting and analyzing viral protein conformations, with a focus on veterinary viruses. The methods covered include homology modeling, molecular dynamics simulations, and machine learning-based structure prediction (e.g., AlphaFold). Their applications in antiviral drug design and vaccine epitope selection are discussed, along with the essential databases and software tools that support this work. The goal is to provide a framework for how these structural and dynamic insights can be translated into viable antiviral strategies for animal health.

2. Computational Methods for Viral Protein Structure Prediction

2.1 Homology Modeling and Template-Based Approaches

Homology modeling, also known as comparative modeling, remains a widely used method for constructing three-dimensional models of viral proteins when experimental structures are unavailable [8, 29]. This approach relies on the principle that protein sequences with significant evolutionary relatedness share conserved three-dimensional folds [8]. The process involves identifying a suitable template structure from the Protein Data Bank (PDB), aligning the target sequence to the template, building the model, and refining it through energy minimization [8].

The accuracy of homology models depends heavily on the sequence identity between the target and template. For viral proteins, templates are often derived from related viral family members. For example, studies of Zika virus (ZIKV) envelope protein have utilized templates from other flaviviruses to model conformational changes associated with glycan loop deletions [30]. Similarly, modeling of host receptor variants, such as angiotensin-converting enzyme 2 (ACE2) allelic forms, has been performed using known crystal structures of the ACE2-spike protein complex to predict variant impacts on binding affinity [8]. The application of homology modeling to veterinary pathogens, including PRRSV glycoproteins, has allowed structural hypotheses to be generated before experimental structures are solved [4, 6].

2.2 Machine Learning and Deep Learning: AlphaFold and Beyond

The advent of deep learning-based methods, particularly AlphaFold, has revolutionized protein structure prediction [1]. These tools predict three-dimensional coordinates directly from amino acid sequences with unprecedented accuracy, often performing at levels comparable to experimental methods for single-domain proteins [1]. For viral proteins, AlphaFold has been used to model envelope glycoproteins, proteases, polymerases, and non-structural proteins from diverse viruses, including coronaviruses, flaviviruses, and bunyaviruses [9, 10, 11, 12, 1].

Machine learning models are particularly valuable for viral targets that are difficult to crystallize or study by cryo-electron microscopy (cryo-EM) due to their inherent flexibility or low abundance. For example, deep learning has been applied to predict the structures of viral methyltransferases and helicases, providing targets for inhibitor design [9]. The integration of AlphaFold predictions with downstream molecular dynamics simulations allows for the exploration of conformational ensembles beyond the predicted static structure [7, 13]. This combined approach is increasingly used in veterinary virology to assess the impact of mutations in emerging viral variants, such as those in PRRSV non-structural proteins linked to recombination hotspots [5].

2.3 Experimental Structure Determination and Its Computational Integration

Experimental methods remain the gold standard for viral protein structure determination [1, 2, 3, 28]. X-ray crystallography provides high-resolution atomic details of stable protein conformations, as demonstrated for Zika virus envelope protein crystals [30] and for engineered coronavirus spike proteins [35]. Nuclear magnetic resonance (NMR) spectroscopy offers insights into protein dynamics in solution, which is particularly useful for smaller viral domains such as nucleoprotein domains [10, 1]. Cryo-electron microscopy (cryo-EM) and cryo-electron tomography (cryo-ET) have become indispensable for visualizing large viral complexes, including the intact SARS-CoV-2 spike trimer and the hantavirus replication machinery, in multiple conformational states [1, 2, 3, 35].

Computational tools are essential for processing cryo-EM data (e.g., using packages such as Relion or cryoSPARC), building atomic models into electron density maps, and validating those models [1, 3]. The integration of experimental maps with computational modeling, including flexible fitting using MD simulations, enables the construction of dynamic models of viral entry and assembly [2, 29].

The table below summarizes the principal methods for viral protein structure determination and prediction.

| Method | Spatial Resolution | Dynamic Information | Applicability to Veterinary Viruses | Key References | |, - |, - |, - |, - |, - | | X-ray Crystallography | High (atomic) | Low (static) | Mature platforms; capsid and envelope proteins | [1, 3, 30, 35] | | Cryo-EM / Cryo-ET | Near-atomic to medium | Medium (multiple states) | Large complexes; virion architecture | [1, 2, 3] | | NMR Spectroscopy | High (atomic) | High (solution dynamics) | Small domains; intrinsically disordered regions | [10, 1] | | Homology Modeling | Template-dependent | Low | Broad; emerging viruses with family templates | [8, 29] | | Deep Learning (AlphaFold) | High (predicted) | Low (static prediction) | Any sequenced virus; blind prediction | [9, 10, 11, 12, 1] |

3. Molecular Dynamics Simulations of Viral Proteins

3.1 Principles and Setup

Molecular dynamics simulations provide a computational microscope for observing the time-dependent behavior of viral proteins at atomic resolution [29, 35]. The method solves Newton's equations of motion for a system of interacting atoms using a molecular mechanics force field (e.g., CHARMM, AMBER, or GROMACS force fields) [29]. For viral simulations, typical setups include solvating the protein in a water box, adding ions to neutralize the system, and running energy minimization followed by equilibration phases (NVT and NPT ensembles) before the production run [29].

The choice of force field and simulation parameters is critical for capturing biologically relevant conformations. For viral glycoproteins, which contain post-translational modifications such as glycosylation, specialized force field parameters for carbohydrates are required [29, 35]. Simulations of viral envelope proteins often span tens to hundreds of microseconds to observe conformational transitions or ligand binding events [29]. These timescales, though computationally expensive, are accessible through specialized hardware (e.g., GPU clusters) or coarse-grained models.

3.2 Applications to Viral Envelope Glycoprotein Dynamics

MD simulations have been extensively applied to study the conformational dynamics of viral fusion proteins. For coronaviruses, simulations of the spike protein have elucidated the mechanism of receptor-binding domain (RBD) opening and closure, transitions that are essential for host cell recognition [32, 35]. Simulations have also characterized the stabilizing effect of engineered disulfide bonds designed to lock the spike in the prefusion conformation for vaccine development [35]. Similar studies have been conducted on the hemagglutinin of influenza viruses and the glycoproteins of flaviviruses, revealing how pH changes and receptor binding trigger the fusogenic transition [29, 33, 34].

For veterinary pathogens, MD simulations have been used to study the GP4 envelope protein of PRRSV and its interaction with neutralizing antibodies [4]. Simulations of FMDV capsid proteins have provided insights into the stability of the virion under different environmental conditions [31]. The dynamic behavior of the Dengue virus envelope has been shown to change at physiological temperatures, affecting antibody recognition [33, 34]. These studies highlight the utility of MD simulations in predicting epitope exposure and designing immunogens that present conserved neutralizing epitopes.

3.3 Free Energy Calculations and Binding Affinity Prediction

A key application of MD simulations is the calculation of binding free energies between viral proteins and potential antiviral compounds or antibodies [9, 14]. Methods such as Molecular Mechanics Poisson-Boltzmann Surface Area (MM-PBSA) and Molecular Mechanics Generalized Born Surface Area (MM-GBSA) provide estimates of binding affinity from simulation trajectories [9]. Free energy perturbation (FEP) and thermodynamic integration offer higher accuracy for predicting the relative binding affinities of related ligands [29].

These computational tools have been used to identify small-molecule inhibitors targeting viral proteases and polymerases. For instance, in silico screening combined with MD-based binding energy calculations has identified gallic acid as a potential inhibitor of Zika virus NS3 and RNA-dependent RNA polymerase (RdRp) [14]. Similarly, structure-based virtual screening has been applied to the Dengue virus NS5 protein to identify novel druggable sites suitable for pan-serotype antiviral design [9]. These approaches are directly transferable to veterinary pathogens, such as the 3CL protease of porcine coronaviruses [15].

3.4 Simulation of Non-Structural Proteins and RNA Elements

Beyond envelope proteins, MD simulations are valuable for studying viral non-structural proteins and regulatory RNA elements. The macrodomain 1 of SARS-CoV-2, which possesses ADP-ribosyl hydrolase activity, has been studied via computational modeling to identify host interaction partners [16]. Simulations of the SARS-CoV-2 frameshifting element (FSE), a structured RNA region, have revealed the role of alternative conformations in programmed ribosomal frameshifting, a process essential for viral gene expression [17, 13]. For veterinary viruses, similar RNA structural elements in the PRRSV NSP9 gene have been identified as recombination hotspots, and their stability can be predicted using RNA folding algorithms [5].

The following Mermaid diagram illustrates a typical computational workflow for predicting viral protein conformations for antiviral design.

flowchart TD
    A[Viral Sequence Retrieval & Annotation], > B[Template Identification / AlphaFold Prediction]
    B, > C{Structure Quality Check}
    C, Good, > D[Molecular Dynamics Setup]
    C, Poor, > E[Experimental Structure / Cryo-EM Refinement]
    E, > D
    D, > F[Equilibration & Production MD]
    F, > G[Conformational Analysis & Clustering]
    G, > H[Binding Site Identification / Cryptic Pocket Detection]
    H, > I[Virtual Screening / Free Energy Calculation]
    I, > J[Lead Compound Selection & Validation]
    J, > K[In Vitro / In Vivo Antiviral Testing]

4. Structural Databases and Software Ecosystems

4.1 The Protein Data Bank (PDB) and UniProt

The Protein Data Bank (PDB) is the primary repository for experimentally determined macromolecular structures [1, 3, 35]. For veterinary virology, the PDB contains structures of viral capsid proteins, envelope glycoproteins, polymerases, and host receptor complexes. The UniProt Knowledgebase provides comprehensive sequence and functional annotation for viral proteins, which is the starting point for any structural modeling project. Cross-referencing PDB entries with UniProt allows for the selection of high-quality templates for homology modeling and validation of predicted models.

For emerging viruses with no experimental structure, the PDB still serves as a source of homologous templates from related viral families. For example, the structures of flavivirus envelope proteins from Dengue and Zika viruses have been used to model the corresponding proteins in lesser-studied viruses such as Yezo virus [10, 2, 28].

4.2 Key Computational Software Packages

Several software packages are essential for structural virology and molecular dynamics workflows. The Rosetta suite is widely used for protein structure prediction, design, and docking [1]. It can perform ab initio modeling of small domains, protein-protein docking, and loop refinement. A core application is the prediction of how mutations affect protein stability and binding.

GROMACS is one of the most popular open-source packages for molecular dynamics simulations [29]. Its high performance on GPU clusters makes it suitable for simulating large viral systems. Other widely used packages include AMBER, NAMD, and CHARMM.

For machine learning-based structure prediction, AlphaFold2 and its derivatives (e.g., AlphaFold3) have become standard tools [1]. These tools require minimal computational expertise to run and produce models that can be directly used for further analysis, including docking studies and MD simulations.

4.3 Practical Workflow Integration

The typical workflow for antiviral target discovery in veterinary medicine integrates these tools in a stepwise manner. Genomic sequencing of field isolates provides the viral protein sequences. Homology modeling or AlphaFold is used to generate initial three-dimensional models. These models are subjected to MD simulations to explore conformational flexibility and identify druggable pockets. Virtual screening, using docking programs such as AutoDock Vina or Glide, identifies candidate small molecules for experimental testing. Free energy calculations refine these predictions for the most promising compounds.

For vaccine design, the workflow focuses on identifying epitopes. MD simulations can predict which regions of a viral protein are solvent-exposed and conformationally stable, making them good vaccine targets [18, 12, 19]. For example, reverse vaccinology approaches for monkeypox virus have combined pan-genome analysis with structural prediction to design multi-epitope vaccines [12]. Similar strategies are applied to Dengue virus and PRRSV, where serotype-specific and conserved epitopes are identified using computational methods [4, 19, 6].

5. Applications in Antiviral Drug Design and Vaccine Development

5.1 Targeting Viral Proteases and Polymerases

Viral proteases and polymerases are established targets for antiviral drug development [15, 9, 14]. Their structures are often highly conserved among related viruses, making them attractive targets for broad-spectrum inhibitors. For coronaviruses, the main protease (3CLpro) has been the focus of extensive structure-based drug design, with computational methods exploring chemical space to design inhibitors with activity against multiple coronavirus species [15]. For flaviviruses, the NS5 protein, which contains both a methyltransferase and an RNA-dependent RNA polymerase, has yielded novel druggable sites identified through comprehensive in silico exploration [9].

In veterinary medicine, these approaches are directly applicable to pathogens such as PRRSV and FMDV. The 3C-like protease of FMDV is a validated target, and computational screening against its active site has the potential to identify novel antiviral compounds [31]. Similarly, the RdRp of PRRSV is a promising target for the design of nucleoside analog inhibitors.

5.2 Design of Entry Inhibitors and Fusion Inhibitors

Viral envelope proteins mediating host cell entry are prime targets for neutralizing antibodies and fusion inhibitors [20, 32, 35]. Structure-guided design of peptide-based fusion inhibitors has been successful for class I fusion proteins, such as those found in coronaviruses, influenza, and paramyxoviruses [20]. The computational prediction of the interaction between viral fusion cores and host receptors (e.g., TLR4/MD-2) allows for rational design of decoy peptides that block membrane fusion [20].

For veterinary viruses, such as PRRSV, the GP2, GP3, GP4, and GP5 envelope proteins are essential for host cell entry and are being targeted for vaccine and inhibitor design [4, 6]. Structural modeling of these glycoproteins in complex with monoclonal antibodies has elucidated the epitopes responsible for neutralization, providing a template for immunogen design [4].

5.3 Epitope Mapping and Vaccine Design

Computational epitope prediction is a critical step in reverse vaccinology [18, 12, 19]. B-cell epitope prediction algorithms, combined with structural filtering to identify surface-exposed regions, reduce the number of candidate peptides that need experimental validation [19]. For example, an integrated in silico approach has been used to identify serotype-specific B-cell epitopes from Dengue virus non-structural proteins, which could aid in serodiagnosis and vaccine design [19]. For SARS-CoV-2 and MERS-CoV, rational design of ferritin nanoparticle vaccines displaying optimized spike protein epitopes has been guided by structural and computational analysis [18].

In the veterinary context, similar strategies are used to develop vaccines for PRRSV, FMDV, and avian influenza. The identification of conserved epitopes across different strains is particularly important for designing broadly protective vaccines. Molecular dynamics simulations of these epitopes can assess their stability and immunogenicity before experimental testing [12, 33].

6. Challenges and Future Directions

Despite the power of computational methods, several challenges remain. High mutation rates in RNA viruses, such as those causing influenza and PRRSV, lead to rapid antigenic drift, requiring iterative structural modeling and sequence surveillance [7, 30]. The computational prediction of the fitness consequences of non-spike mutations, as seen in emerging SARS-CoV-2 variants, requires analysis of the entire proteome rather than focusing on surface proteins alone [7].

The integration of experimental data from cryo-EM with MD simulations is an area of active development. Hybrid methods, such as flexible fitting and metainference, allow for the construction of dynamic models that are consistent with experimental density maps [3, 29]. The development of coarse-grained models permits simulations of entire virions or large replication complexes, expanding the timescales accessible to computation.

Future advances in machine learning, including foundation models that can predict protein dynamics directly from sequence, promise to further accelerate the discovery of antiviral targets [1]. The application of these approaches to the diverse array of viruses affecting animal health remains a critical frontier.

7. Conclusion

The integration of structural virology and molecular dynamics provides a powerful framework for predicting viral protein conformations central to antiviral and vaccine design. Computational methods, from homology modeling to deep learning and atomistic simulations, enable the characterization of viral protein dynamics that static experimental structures cannot capture. These approaches are directly applicable to major veterinary pathogens, including PRRSV, FMDV, coronaviruses, and flaviviruses. By linking structural data with functional predictions, researchers can rationally design inhibitors, map neutralizing epitopes, and prepare for emerging zoonotic threats in animal populations.

References

[1] Tanuj Handa, Ankita Saha, Aarthi Narayanan et al. Structural Virology: The Key Determinants in Development of Antiviral Therapeutics. Viruses. 2025. URL: https://www.semanticscholar.org/paper/d58496f79293b08cfb2b67385ef5b425de2a5e90

[2] Kristina Meier, S. Thorkelsson, E. Quemin et al. Hantavirus Replication Cycle, An Updated Structural Virology Perspective. Viruses. 2021. URL: https://www.semanticscholar.org/paper/b7cbc6b8572d6bf94ebdfe85689a5eb88e8d7944

[3] Nathan Hardenbrook, Peijun Zhang. A structural view of the SARS-CoV-2 virus and its assembly. Current Opinion in Virology. 2021. URL: https://www.semantics

[4] Veselkova B, Saikia C, Affeldt S et al. Structural basis of porcine reproductive and respiratory syndrome virus 2 neutralization by a GP4-targeting monoclonal antibody. J Gen Virol. 2026. URL: https://pubmed.ncbi.nlm.nih.gov/42348291/

[5] Wang X, Pang Y, Chen Y et al. Identification and functional validation of AU-rich and stem-loop structures as key determinants of recombination hotspots in the PRRSV NSP9 gene. J Gen Virol. 2026. URL: https://pubmed.ncbi.nlm.nih.gov/42340752/

[6] Pinto D, Rowland RRR, Brandariz-Nuñez A. Signal peptide cleavage and ectodomain regions of GP2 are required for PRRSV infection. J Gen Virol. 2026. URL: https://pubmed.ncbi.nlm.nih.gov/42313464/

[7] Shuhei Tsujino, Masumi Tsuda, Naganori Nao et al. Evolution of BA.2.86 to JN.1 reveals that functional changes in non-structural viral proteins are required for fitness of SARS-CoV-2. Journal of Virology. 2025. URL: https://www.semanticscholar.org/paper/d7562b73e27ab36f5eca331eeaaf83a41c170481

[8] Mushtaq Hussain, N. Jabeen, Fozia Raza et al. Structural variations in human ACE2 may influence its binding with SARS-CoV-2 spike protein. Journal of Medical Virology. 2020. URL: https://www.semanticscholar.org/paper/90103d25d445809d530622f1b88a61bcf652a384

[9] Trigueiro-Louro J, Correia V, Ali IDS et al. Dengue Virus NS5 Target Discovery: A Comprehensive in Silico Exploration of Novel Druggable Sites for Pan-Serotype Antiviral Design. Int J Mol Sci. 2026. URL: https://pubmed.ncbi.nlm.nih.gov/42353353/

[10] Gladysheva AV, Yanshin AO, Radchenko NS et al. Solution Structure of Nucleoprotein Domain 1 from the Emerging Yezo Virus. Int J Mol Sci. 2026. URL: https://pubmed.ncbi.nlm.nih.gov/42353207/

[11] Archana K, Agrahari A, Mishra S et al. Identification of JEV capsid interacting domain(s) with host protein(s): Novel target for developing anti-JEV therapeutics. Life Sci. 2026. URL: https://pubmed.ncbi.nlm.nih.gov/42342069/

[12] Saber MA, Shovon MHJ, Imtiaz M et al. Pan-genome and reverse vaccinology for a multi-epitope vaccine against circulating post-2022 Monkeypox virus strains. PLoS One. 2026. URL: https://pubmed.ncbi.nlm.nih.gov/42330028/

[13] Abhishek Dey, Shuting Yan, Tamar Schlick et al. Abolished frameshifting for predicted structure-stabilizing SARS-CoV-2 mutants: implications to alternative conformations and their statistical structural analyses. bioRxiv. 2024. URL: https://www.semanticscholar.org/paper/b6bf5e7b856b809c2cfb441d7f2a1d11eaa5556f

[14] Heo JH, Kang N, Lee YJ et al. In Silico Prediction of Gallic Acid Binding to Zika Virus NS3/RdRp Proteins and In Vitro Confirmation of Its Antiviral Activity. Eur J Pharmacol. 2026. URL: https://pubmed.ncbi.nlm.nih.gov/42364892/

[15] Gu Z, Qi Z, Chen Z et al. Exploring chemical space based on Transformation to design broad-spectrum 3CL(pro) inhibitors against coronavirus. Eur J Med Chem. 2026. URL: https://pubmed.ncbi.nlm.nih.gov/42364457/

[16] Cameron CD, Heilmann G, Roman BK et al. Interactomics of SARS-CoV-2 Macrodomain 1 Reveals Putative Clients of ADP-Ribosyl Hydrolase Activity. Viruses. 2026. URL: https://pubmed.ncbi.nlm.nih.gov/42357598/

[17] Ross SJ, Ye C, Moxon S et al. RNA Folding Energy of Long-Range Genomic Interactions Regulates Discontinuous Transcription in SARS-CoV-2. Viruses. 2026. URL: https://pubmed.ncbi.nlm.nih.gov/42357630/

[18] Liu Y, Rao H, Wang N et al. Rational design of an optimized ferritin nanoparticle vaccine targeting both SARS-CoV-2 and MERS-CoV. J Nanobiotechnology. 2026. URL: https://pubmed.ncbi.nlm.nih.gov/42363226/

[19] Subramanian P, Shanmugaraj B, Ramalingam S. Identification of candidate serotype-specific B-cell epitopes from dengue virus non-structural proteins using an integrated in silico approach. Arch Virol. 2026. URL: https://pubmed.ncbi.nlm.nih.gov/42323433/

[20] Kircheis R. In Silico Modeling of Structural Compatibility and Alignment Between Viral Class I Fusion Cores and Human TLR4/MD-2. Int J Mol Sci. 2026. URL: https://pubmed.ncbi.nlm.nih.gov/42353037/

[21] Xu W, Gonepudi NK, Han Y et al. Genetic code expansion for single noncanonical amino acid incorporation into SARS-CoV-2 Omicron spike on amber-free lentiviral particles. Methods Enzymol. 2026. URL: https://pubmed.ncbi.nlm.nih.gov/42349953/

[22] Yang RH, Gao WC, Zhang L et al. ZIKV NS3 disrupts Nrf2-mediated antioxidant defense and drives vascular smooth muscle cell damage and vascular dysfunction. Zool Res. 2026. URL: https://pubmed.ncbi.nlm.nih.gov/42339736/

[23] Xiao Y, Liu Z, Hu S et al. Unveiling the molecular mechanism of Qingwen Baidu decoction against dengue fever: an integrated study of bioinformatic analysis, machine learning and network pharmacology. Funct Integr Genomics. 2026. URL: https://pubmed.ncbi.nlm.nih.gov/42332327/

[24] Shinde RV, Mohite ST, Durgawale PP et al. History, Virology, Pathogenesis, Epidemiology, and Laboratory Diagnosis of Dengue Virus: A Focused Review. Cureus. 2026. URL: https://pubmed.ncbi.nlm.nih.gov/42326244/

[25] Medits-Weiss I, Roßbacher L, Springer DN et al. Conserved histidine residues and the control of tick-borne encephalitis virus maturation. J Gen Virol. 2026. URL: https://pubmed.ncbi.nlm.nih.gov/42313462/

[26] Chuquimia Valdez JA, Guimarães NR, Fonseca V et al. Detection of and Early Genomic Insights into Chikungunya Virus, Bolivia, 2025. Emerg Infect Dis. 2026. URL: https://pubmed.ncbi.nlm.nih.gov/42366174/

[27] S. Bos, Wildriss Viranaicken, Jonathan Turpin et al. Corrigendum to "The structural proteins of epidemic and historical strains of Zika virus differ in their ability to initiate viral infection in human host cells". Virology. 2019. URL: https://www.semanticscholar.org/paper/81689827d14925d86257c9adee54734702d3a824