Yongzheng Hu
Wei Jiang- Department of Nephrology, The Affiliated Hospital of Qingdao University, Qingdao, Shandong, China
Chronic hepatitis B virus (HBV) persists through a compact proteome, deep reliance on host pathways, and a nuclear covalently closed circular DNA (cccDNA) reservoir that current antivirals rarely extinguish. This Mini Review synthesizes advances from 2020–2025 in which artificial intelligence (AI) augments mechanistic understanding of HBV rather than serving only predictive ends. We summarize (i) AI-enabled structural modeling that clarifies polymerase priming and HBx architecture; (ii) physics-informed and multiscale inference that links sparse measurements to replication and cccDNA kinetics; (iii) sequence-based learners that expose non-random host-genome integration contexts and mutational constellations associated with immune tolerance or escape; (iv) network-aware analyses that prioritize host dependencies and connect CRISPR perturbations to virus–host modules governing cccDNA transcriptional control; and (v) AI-assisted antiviral discovery that couples virtual screening with mechanism-anchored interpretation (e.g., capsid assembly modulators). Across these domains, AI sharpens hypotheses by mapping viral mutations and host factors to discrete steps of the life cycle, quantitatively elevating high-leverage processes such as nucleocapsid recycling and cccDNA silencing, and guiding structure- or phenotype-guided intervention design. We also outline practical constraints—data sparsity, cross-cohort heterogeneity, and interpretability—and propose priorities that couple computation and experiment: mechanism-aware gray-box models, causal and spatially resolved analyses, calibrated uncertainty and benchmarking across genotypes, and active-learning loops that maximize information gained per experiment. Framed this way, AI emerges as a mechanism-aware partner to experimental virology, accelerating routes toward durable functional cure through eradication or stable transcriptional silencing of cccDNA.
Graphical Abstract.
Introduction
Hepatitis B virus (HBV) chronically infects over 250 million people worldwide and causes significant morbidity and mortality due to liver cirrhosis and hepatocellular carcinoma (2, 3). The persistence of HBV is primarily driven by the formation of covalently closed circular DNA (cccDNA) in the nuclei of infected hepatocytes (4), which serves as a stable episomal mini-chromosome that continuously produces viral RNAs and progeny virions (5). This nuclear cccDNA reservoir is not eliminated by current antiviral therapies and is a key obstacle to curing chronic HBV infection. Additionally, integration of HBV DNA into the host genome can contribute to oncogenesis and HBsAg production, further complicating disease persistence (6, 7). HBV’s life cycle relies on intricate interactions between viral proteins and host cell factors: for example, the HBx protein modulates epigenetic and signaling pathways to favor viral persistence, and the viral polymerase (Pol) carries unique domains for genome replication that have proven challenging to characterize structurally (8, 9). These and other mechanistic complexities – including immune evasion via exhausted T cell responses and decoy viral antigens – underscore the need for new approaches to dissect HBV biology (10).
In parallel, the past decade has seen rapid growth in the application of artificial intelligence (AI) and machine learning (ML) in biomedical research (11, 12). AI methods excel at finding patterns in high-dimensional data and have already revolutionized fields like protein structure prediction and image analysis. In virology, AI approaches have been employed to study virus–host interactions, predict viral evolution, and improve antiviral drug discovery (13–15). For HBV, which generates diverse data from genomic sequences to patient biomarkers, AI offers a powerful toolkit to uncover hidden relationships and generate predictive models that complement experimental findings. Crucially, AI techniques can help bridge gaps where traditional virological methods face limitations: for instance, predicting viral behavior without invasive biopsies, or simulating molecular processes that are difficult to capture experimentally.
Here, we delineate how AI has advanced mechanistic understanding of HBV, emphasizing applications from roughly 2020 to 2025 that interrogate replication and cccDNA kinetics, resolve viral protein structure–function (polymerase, HBx), map host–virus dependency networks, and parse immune modulation relevant to persistence. Our synthesis privileges studies in which AI yields explanatory insight—pinpointing viral or host determinants that govern cccDNA maintenance, polymerase activity, capsid dynamics, or immune escape—rather than purely prognostic models unless they illuminate underlying biology. We further identify methodological and data constraints and outline where computation and experiment should be coupled to address HBV-specific bottlenecks. Framed this way, AI emerges not as an adjunct but as a mechanism-aware partner to experimental virology, sharpening hypotheses, prioritizing interventions, and accelerating routes toward durable functional cure.
To orient readers across methods and mechanistic endpoints, Table 1 maps representative AI approaches to core HBV questions—structure/function, replication and cccDNA kinetics, genomics/integration, host–virus networks, and antiviral discovery.
Table 1. AI approaches mapped to HBV mechanistic questions.
Selecting an AI framework should reflect inductive bias, data regime, and validation. Physics-informed networks embed governing Ordinary Differential Equation/Partial Differential Equations (ODE/PDEs) into the loss (32), constraining fits to biologically plausible trajectories and enabling parameter identification under sparse time-series (e.g., latent cccDNA kinetics); they differ from neural ODEs by enforcing explicit mechanistic residuals rather than learning unconstrained latent dynamics. PINNs thus excel when priors are strong and labels are limited, but require credible process models and careful scaling of data–physics terms. In contrast, attention-based sequence learners/Transformers capture long-range dependencies in genomes and epigenomic context (33), well-suited for HBV integration hotspot discovery and motif attribution (24), but need larger labeled sets and rigorous interpretability checks. Structure predictors (e.g., AlphaFold-class models) leverage evolutionary and geometric priors to generate atomistic hypotheses for Pol/HBx that can be tested by mutagenesis or docking (16). Across paradigms, validation should include internal resampling plus external validation on independent cohorts/systems, and whenever possible orthogonal experimental tests (e.g., cryo-EM comparison, CRISPR perturbation, targeted biochemistry). This pragmatic rubric—mechanistic bias vs. data scale, with tiered validation—helps match models to HBV questions spanning replication kinetics, integration, and host-dependency mapping.
AI-driven structural insights into HBV proteins
A central barrier in HBV biology has been the lack of high-confidence structures for key viral proteins (34, 35). Deep learning–based folding (e.g., AlphaFold2) recently produced the first full-length model of HBV polymerase (Pol), resolving a decades-long impasse (17). The model reveals an unexpected arrangement in which the terminal protein (TP) wraps around the reverse transcriptase (RT), positioning the priming tyrosine directly above the RT active site—providing a structural rationale for protein-primed DNA synthesis. RT and RNase H adopt a retrovirus-like orientation despite limited sequence homology, and the overall fold, including TP placement, appears conserved across mammalian and avian hepadnaviruses. These insights rationalize prior mutant phenotypes, enable in silico docking, and open structure-guided avenues beyond the active site (e.g., allosteric pockets), advancing mechanistic studies of replication and cccDNA generation.
AI modeling has likewise suggested HBx, a small, partially disordered regulator essential for replication and implicated in oncogenesis (36). AlphaFold yields a coherent HBx fold with similarity to palmitoleoyl transferases, suggesting membrane-proximal functions or functional mimicry (18). Comparison with alternative threading models and integration with partial experimental data and molecular dynamics favor the AlphaFold solution. Conservation analysis highlights residues critical for transcriptional activity (e.g., Cys61, Cys69, Cys137, His139) across hepadnaviruses, alongside human-HBV-specific sites linked to subcellular targeting, implying a conserved core with host-adapted features. These predictions generate testable hypotheses for HBx–host interactions (e.g., DDB1 engagement, membrane association) (37).
Beyond Pol and HBx, AI-enabled structural virology supports hypothesis-driven work on other proteins. For HBcAg and HBsAg—already characterized by crystallography/cryo-EM (38)—AI can assess mutation impacts, heterogeneity, and conformational preferences, refining genotype–phenotype mapping and antigenicity models. Notably, the Pol structure is already informing interpretation of deep mutational scanning by mapping functional effects onto a 3D framework. More broadly, coupling AI-predicted structures with molecular dynamics should illuminate functionally relevant conformational changes during reverse transcription and capsid assembly (19, 39). Collectively, these advances move previously “undruggable” targets into view and provide a structural foundation for mechanistic experimentation and antiviral design.
AI in modeling HBV replication and cccDNA dynamics
AI is extending classical ODE‐based frameworks of HBV dynamics to enable data-efficient inference for hard-to-observe compartments such as cccDNA and intracellular capsids (40, 41). Physics-informed neural networks embed established virological equations into the loss function so that fitted solutions respect known kinetics while accommodating sparse, noisy time series. A recent virology-informed neural network (VINN) instantiated a four-compartment model (susceptible hepatocytes, infected hepatocytes, rcDNA-containing capsids, and free virus) with explicit capsid recycling to replenish cccDNA, and recovered infection, production, and cccDNA replenishment rates even when only capsid measurements were available (20). The resulting parameter sets were biologically plausible and supported in silico counterfactuals (e.g., varying cccDNA degradation) for mechanism-guided scenario testing.
Physics-informed neural networks (PINNs) differ from neural ODEs in how they encode mechanism and handle sparse data (21, 22). PINNs train the network while penalizing violations of known life-cycle equations and initial/boundary conditions, so the learned trajectories stay close to biologically plausible HBV dynamics; this tends to improve parameter identifiability under limited, noisy time-series and helps recover latent compartments such as cccDNA or intracellular capsids—provided the process equations are credible and the data/physics terms are balanced appropriately. By contrast, neural ODEs learn a flexible latent vector field from data with minimal mechanistic constraints (23); they are useful when prior knowledge is weak or measurements are rich, but may require extra regularization if one needs properties like mass balance or saturation to hold. In practice, many HBV use-cases benefit from a “gray-box” compromise: embed the trusted parts of the replication model (e.g., infection, production, recycling) and let a small learnable residual capture unmodeled effects, then validate with held-out time windows/genotypes and orthogonal assays in addition to predictive metrics. This rubric—mechanistic bias vs. data regime, with tiered validation—helps non-specialists choose an appropriate framework for replication and cccDNA kinetics.
Complementary multiscale models connect intrahepatic HBV DNA/cccDNA kinetics to serum biomarkers (HBsAg, HBV DNA, HBcrAg), with machine learning optimizing fits across large clinical datasets (42, 43). These frameworks can noninvasively estimate intrahepatic cccDNA trajectories, recapitulate the slow decline observed under nucleos(t)ide analog therapy, and reveal substantial inter-patient heterogeneity in cccDNA turnover—implicating host or viral determinants that AI can help identify through feature discovery (44).
Model-based sensitivity analyses converge on two dominant levers—cccDNA replenishment via capsid recycling and cccDNA decay—thereby defining tractable intervention points. By capturing high-order coupling, AI models convert these rankings into mechanism-driven experiments rather than purely descriptive fits.
Machine learning for HBV genomics and integration
The 3.2-kb HBV genome is highly diverse across 9–10 genotypes and myriad variants, with specific mutations shaping replication efficiency, immune escape, and oncogenic risk (25, 45–47). AI has been leveraged to extract mechanistic signals from these sequences. A leading example is deep learning for host-genome integration: an attention-based model (DeepHBV) trained on curated integration loci discriminates true sites from genomic background by local sequence context and auxiliary features (24). Incorporating repetitive elements and cancer epigenomic signals markedly boosts performance (AUC rising from ~0.64 with sequence alone to ~0.94 with full features), indicating that integration strongly favors particular chromatin states and repeat architectures. Genome-wide deployment of the trained model nominates previously unreported hotspots, while attention weights highlight candidate sequence motifs (e.g., short direct repeats) that plausibly scaffold microhomology-mediated integration—mechanistic clues with direct relevance to HCC pathogenesis and surveillance.
Machine learning has likewise linked viral genomic variation to immune phase and control. Classifiers trained on whole-genome sequences from chronic infection cohorts separate immune-tolerant from immune-active HBeAg-positive cases with ~0.88 AUC, outperforming conventional markers by exploiting mutational and diversity features (48). These models “rediscover” canonical determinants—such as the precore stop codon (G1896A) and basal core promoter substitutions associated with HBeAg loss—while suggesting additional combinatorial signatures that modulate whether replication proceeds quietly or provokes immune attack, thereby framing testable hypotheses on epitope presentation, RNA structure, and transcriptional regulation.
Together, these approaches tighten genotype–phenotype links in HBV by exposing DNA-level determinants of integration and by associating specific mutational constellations with immunological state and treatment responsiveness.
AI and HBV host–virus interactions
HBV commandeers host processes throughout its cycle—NTCP-mediated entry, cccDNA maintenance, host-driven transcription, assembly, and egress—while regulators such as HBx remodel DDB1-dependent proteostasis, epigenetic control, and innate sensing (26, 27). AI now systematizes this landscape. Deep learning on omic datasets resolves host-response states and prioritizes dependency factors (28); integrative network models that fuse PPI maps with CRISPR screens consistently surface SMC5/6, proteasome function, calcium signaling, and chromatin modifiers governing cccDNA (29, 49, 50). Features extracted from infection transcriptomes implicate RNA modification machinery (e.g., m^6A writers/readers) as pro-/antiviral regulators, subsequently validated by perturbation of HBV transcripts (51).
Immune interplay is similarly tractable. Neural models for T- and B-cell epitope prediction scan HBV proteins across genotypes to nominate peptides with high HLA class I/II presentation or antibody accessibility and to flag escape-associated substitutions in core or polymerase, complementing experimental evidence of CTL evasion and informing therapeutic immunogen design tuned to population HLA diversity (52). Single-cell and repertoire analyses aided by ML further delineate exhaustion phenotypes and cytokine programs associated with viral control versus persistence (30, 53).
Phenotypic readouts add orthogonal evidence. High-content microscopy combined with convolutional classifiers distinguishes infection stages from fluorescence signatures of viral markers, enabling automated assessment of entry, replication, nuclear core trafficking, assembly, and release; detected phenotypes (e.g., nuclear core accumulation) direct targeted biochemical studies (31, 54). Emerging computational pathology quantifies spatial relationships between infected hepatocytes and immune infiltrates, linking microenvironmental architecture to immune containment or escape. Collectively, these AI-enabled approaches compress high-dimensional host–virus data into testable mechanisms and convergent pathways that HBV exploits, providing a rational basis for host-targeted antivirals and combination strategies that both restrict replication and restore antiviral immunity.
AI in antiviral discovery and therapeutics for HBV
AI is accelerating HBV therapeutic discovery by compressing hit identification and illuminating mechanisms of action. Ligand-based deep learning now predicts antiviral activity directly from molecular structure: a transformer–NN framework (HBCVTr) trained on curated HBV/HCV datasets outperformed traditional QSAR (55), enabled virtual screening of ~10^7 compounds within days, and prioritized novel candidates with docking/MD consistent with core (capsid) protein engagement—thereby shifting brute-force screening toward mechanism-guided triage. Beyond triage, AI helps explain how small molecules perturb viral assemblies. Machine-learning analyses of molecular-dynamics ensembles of HBV core tetramers classify apo versus capsid assembly modulators (CAMs) and separate Class I (accelerating) from Class II (misdirecting) agents with high accuracy, while feature attribution localizes drug-specific “hotspots” on the capsid interface (56). These maps suggest Class I CAMs stabilize early inter-dimer geometries, whereas Class II CAMs induce non-native contacts—actionable insight for designing next-generation modulators and extensible to other targets (e.g., polymerase allostery).
Methodologically, reinforcement learning and knowledge-graph inference are being explored to optimize combination schedules and repurpose compounds that hit HBV dependencies (viral or host). Together, these approaches shorten the path from chemical space to validated leads and, crucially, couple predictive screening with mechanistic interpretability—a prerequisite for rational, multi-pronged regimens that target entry, capsid/replication, cccDNA control, secretion, and immune restoration.
Several AI-guided pipelines have progressed beyond in silico triage toward pre-clinical evaluation. A transformer-QSAR framework (HBCVTr) prioritized core-engaging chemotypes at multi-million–scale and proposed testable hits (e.g., “IJN-04”), establishing a tractable route from sequence-to-activity learning to wet-lab follow-up (55). Complementarily, recent virtual-screening campaigns against HBV core, augmented by ML-assisted docking/MD consensus, yielded novel capsid inhibitors that advanced to in vitro validation, demonstrating HBV replication suppression and acceptable drug-likeness/tox profiles in cell systems (57). While many capsid assembly modulators (CAMs) now in clinical trials were not originally discovered by AI, machine-learning analysis of MD ensembles has clarified their mechanisms—separating Class I (accelerating) from Class II (misdirecting) effects and localizing drug-specific inter-dimer “hotspots”—thereby informing next-generation scaffold design and resistance-risk assessment (56, 58). In parallel, knowledge-graph and QSAR-driven repurposing workflows are maturing, with external validations reported in other viral settings and growing adoption in HBV pipelines to shortlist host/viral targets for bench testing (59, 60). Together, these examples illustrate a practical translation path: AI narrows chemical/target space, MoA-aware models de-risk leads, and cellular assays close the loop en route to animal studies and early clinical evaluation.
Discussion
Recent applications of AI to HBV research illustrate not merely algorithmic novelty but a changing epistemology for mechanistic virology. Where data are scarce, latent, or ethically difficult to obtain, AI has supplied principled surrogates: structural predictions that stand in for recalcitrant crystallography; physics-informed inference that exposes hidden state variables such as cccDNA; and representation learning that compresses high-dimensional host–virus measurements into tractable hypotheses (61–63). The value of these contributions is twofold. First, they expand the space of askable questions—e.g., what domain geometries of polymerase are compatible with protein priming, which genomic contexts are permissive for integration, or which host modules gate transitions between immune tolerance and activity. Second, they translate diffuse signals into experimentally testable statements, enabling a tighter discovery loop between computation and bench work.
A unifying theme is inductive bias: models that encode known virology or physics produce more credible inferences under limited data (64, 65). Physics-informed networks regularize learning with life-cycle constraints (e.g., mass balance, initial/boundary conditions), whereas neural ODEs learn flexible latent dynamics with minimal mechanistic priors; the former is preferable when trusted equations exist, the latter when trajectories are rich but priors are weak (66, 67). Practically, hybrid (“gray-box”) pipelines work well: couple sequence/structure priors with mutational or MD readouts, and augment ODE life-cycle models with a small learnable residual to capture unmodeled effects. This design supports counterfactual tests in silico, and keeps estimates aligned with biology. Applied to HBV, such constrained models consistently elevate the same high-leverage levers—nucleocapsid-mediated cccDNA replenishment and cccDNA loss/silencing—thereby focusing experimental prioritization.
A second practical theme is data governance and benchmarking. HBV datasets are fragmented across platforms (bulk and single-cell transcriptomics, proteomics, imaging, CRISPR perturbations), genotypes, and clinical contexts. Without harmonization, model generalizability and causal interpretability remain uncertain (68, 69). We recommend establishing: (i) genotype-stratified reference corpora with shared metadata and controlled vocabularies (infection system, MOI, time post-infection, readouts); (ii) multi-site challenge tasks with predefined splits for out-of-distribution testing (e.g., training on genotypes B/C, testing on A/D) to quantify portability; (iii) gold-standard targets for cccDNA (paired biopsy/serology time series where available) to calibrate latent-state estimators; and (iv) standardized performance/uncertainty reporting (confidence intervals, calibration curves, decision-relevant metrics such as Net Benefit for triage use-cases).
Causality and perturbation should be foregrounded. Many current studies are associational—useful for hypothesis generation but limited for mechanism (70). HBV is unusually amenable to controlled perturbations (siRNA/CRISPR screens, inducible HBx, CAMs with well-defined MoAs). Embedding such interventional data within causal graphical frameworks and counterfactual learners would sharpen claims about necessity and sufficiency of host factors or viral processes. In practice, marrying pooled-CRISPR readouts with graph neural networks over protein–protein and chromatin contact maps can identify bottleneck nodes whose modulation produces large effects on replication or cccDNA transcription. Importantly, these models should propagate experimental interventions as exogenous “do” operators rather than observational covariates, enabling proper estimation of causal effects.
The spatial dimension of HBV pathobiology is underexploited and well matched to modern AI (71, 72). Spatial transcriptomics/pathology and high-content imaging can map the mesoscale organization of infected hepatocytes, stromal cells, and immune infiltrates. Learning on such data (e.g., attention-based multiple-instance models) can quantify how micro-anatomical niches modulate viral production, antigen presentation, and immune surveillance. Coupling spatial features to VINN-like latent-state models would connect tissue architecture to cccDNA kinetics—potentially explaining patient-to-patient variability in functional cure despite similar systemic exposures.
To be clinically consequential, uncertainty and interpretability must be first-class citizens. For structural models, confidence measures should guide experimental validation (e.g., targeted mutagenesis at low-confidence loops versus high-confidence catalytic motifs) (73). For kinetics and integration predictors, Bayesian or conformal methods can flag low-confidence regimes (rare genotypes, extreme biomarker values), preventing overconfident extrapolation (74, 75). Local attribution (e.g., attention maps in DeepHBV) should be complemented by global analyses (e.g., SHAP interaction values) to uncover non-linear epistasis or motif–chromatin interactions that drive integration. These practices convert “black-box” outputs into actionable priors for the lab.
A forward-leaning agenda should also embrace closed-loop, active learning. In HBV drug discovery, propose–test–update cycles can be formalized so that each experimental batch maximizes expected information gain about a mechanistic objective (e.g., capsid inter-dimer angle distributions, cccDNA transcriptional burst frequency). In host-factor discovery, adaptive CRISPR libraries can refine around emergent network bottlenecks. For immunology, generative models conditioned on HLA distributions and escape landscapes can design peptide sets that are prospectively screened in T-cell assays, with immunogenicity results feeding back into the generator. Across use-cases, this experiment-aware AI can make small datasets perform like large ones by choosing the right measurements (76).
Finally, ethics and dual-use considerations warrant explicit handling, particularly for generative models and high-resolution life-cycle simulators (77, 78). Mechanistic fidelity that empowers cure research could, in principle, be misapplied (79). Governance should include access controls for sensitive models/datasets, transparent model cards documenting training data and intended use, and pre-publication risk assessment. In parallel, privacy-preserving analytics (federated learning, secure aggregation) would allow multi-center HBV cohorts to contribute to robust models without exposing identifiable data—critical for assembling the scale necessary to answer genotype- and ancestry-specific questions. For biosecurity vetting, we align with DURC/P3CO oversight by incorporating pre-publication risk review and institutional vetting of models/datasets most likely to enable misuse, alongside access-controlled release, with model cards that document training data, intended use, and prohibited applications (80, 81). In parallel, upstream controls such as DNA/RNA synthesis customer and sequence screening (e.g., International Gene Synthesis Consortium harmonized protocol) and national measures that strengthen nucleic-acid synthesis procurement screening further reduce misuse pathways for sequence-generating or sequence-editing tools. For data governance, we recommend multi-center collaboration via privacy-preserving analytics—notably federated learning with secure aggregation—to train across institutions without centralized transfer of sensitive data (82), and adoption of GA4GH frameworks/standards (e.g., data-use governance and htsget APIs) to enable auditable access, fine-grained consent, and interoperable sharing of virology datasets. Together, these measures allow the field to scale mechanism-aware AI for HBV while meeting contemporary expectations for biosecurity, privacy, and accountability.
Conclusions and future directions
Artificial intelligence is reshaping HBV mechanistic research by turning scarce, heterogeneous measurements into testable hypotheses about replication, cccDNA persistence, host dependencies, and immune escape. The most durable gains arise when models embed virological structure—physical constraints for proteins, life-cycle kinetics for within-host dynamics, and network priors for host–virus interactions—so that predictions remain biologically interpretable and experimentally actionable.
Future work should prioritize six directions. (i) Mechanism-aware modeling. Advance gray-box approaches that couple ODE/PDE life-cycle models and structural constraints with learnable residuals, enabling counterfactual in silico perturbations on processes such as capsid recycling and cccDNA silencing. (ii) Causal and interventional inference. Integrate CRISPR/perturb-seq and pharmacologic interventions into causal graphs and counterfactual learners to identify necessary and sufficient host dependencies, moving beyond association. (iii) Spatially resolved HBV biology. Fuse spatial transcriptomics/pathology and high-content imaging with latent-state models to connect microanatomy to cccDNA kinetics and immune containment. (iv) Uncertainty and interpretability. Pair Bayesian or conformal calibration with global/local attribution (e.g., interaction-aware feature importance) to delineate confidence regimes, nominate mutation/epitope targets, and guide experimental validation. (v) Benchmarking and data governance. Establish genotype-stratified reference corpora, out-of-distribution challenges across cohorts and genotypes, and shared cccDNA “gold standards,” with standardized reporting of calibration and decision-relevant metrics. (vi) Closed-loop discovery. Use active learning in drug and target discovery—propose, test, and update—to maximize information gain per experiment; extend to immunogen design conditioned on HLA landscapes and escape pathways.
Attention to ethics and dual-use is essential. Access-controlled sharing, transparent model cards, and privacy-preserving analytics (e.g., federated learning) will enable multi-center models without compromising participant data. With these safeguards and priorities, AI can accelerate a shift from descriptive virology to predictive, intervention-ready mechanisms, bringing durable functional cure within closer reach.
Author contributions
YH: Conceptualization, Data curation, Formal Analysis, Methodology, Resources, Software, Validation, Visualization, Writing – original draft. WJ: Conceptualization, Funding acquisition, Investigation, Project administration, Resources, Supervision, Writing – review & editing.
Funding
The author(s) declare financial support was received for the research and/or publication of this article. This research was supported by grants from the National Natural Science Foundation of China (No.82370724), the Qingdao Key Health Discipline Development Fund and the Qingdao Key Clinical Specialty Elite Discipline project.
Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Generative AI statement
The author(s) declare that no Generative AI was used in the creation of this manuscript.
Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.
Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
Abbreviations
AI, artificial intelligence; CAM, capsid assembly modulator; cccDNA, covalently closed circular DNA; CRISPR, clustered regularly interspaced short palindromic repeats; CTL, cytotoxic T lymphocyte; DDB1, damage-specific DNA-binding protein 1; HBcAg, hepatitis B core antigen; HBcrAg, hepatitis B core-related antigen; HBsAg, hepatitis B surface antigen; HBV, hepatitis B virus; HBx, hepatitis B virus X protein; HCV, hepatitis C virus; HLA, human leukocyte antigen; MD, molecular dynamics; ML, machine learning; NTCP, sodium taurocholate cotransporting polypeptide; ODE, ordinary differential equation; PPI, protein–protein interaction; QSAR, quantitative structure–activity relationship; rcDNA, relaxed circular DNA; RNase H, ribonuclease H; RT, reverse transcriptase; SHAP, SHapley Additive exPlanations; VINN, virology-informed neural network.
References
1. Hu Y and Jiang W. Artificial intelligence for mechanistic understanding of hepatitis B virus. Front Virol. (2025).
2. Chen YC, Yang CC, Kuo HT, Sheu MJ, Feng IC, Liu CF, et al. Risk factors and nomogram model for hepatocellular carcinoma development in chronic hepatitis B patients with low-level viremia. Int J Med Sci. (2024) 21:1661–71. doi: 10.7150/IJMS.95861
3. He Q, Chang X, Xu R, Han W, Wang S, Gong S, et al. Global temporal trends and projections of hepatitis B-related cirrhosis among adolescents and young adults from 1990 to 2035: an analysis based on the global burden of disease study 2021. Front Public Health. (2024) 12:1494388/BIBTEX. doi: 10.3389/FPUBH.2024.1494388/BIBTEX
4. Tu T, Zehnder B, Wettengel JM, Zhang H, Coulter S, Ho V, et al. Mitosis of hepatitis B virus-infected cells in vitro results in uninfected daughter cells. JHEP Rep. (2022) 4:100514. doi: 10.1016/J.JHEPR.2022.100514
5. Pastor F, Charles E, Di Vona C, Chapelle M, Rivoire M, Passot G, et al. The dual-specificity kinase DYRK1A interacts with the Hepatitis B virus genome and regulates the production of viral RNA. PloS One. (2024) 19:e0311655. doi: 10.1371/JOURNAL.PONE.0311655
6. Zhang M, Chen H, Liu H, and Tang H. The impact of integrated hepatitis B virus DNA on oncogenesis and antiviral therapy. biomark Res. (2024) 12:1–16. doi: 10.1186/S40364-024-00611-Y/TABLES/2
7. Hui RWH, Wong DKH, Lyu X, Mak LY, Fung J, Seto WK, et al. Profiles of HBV DNA integration in humans with hepatitis B virus infection: Insights for antiviral treatment. JHEP Rep. (2025) 7:101487. doi: 10.1016/J.JHEPR.2025.101487
8. Lun YZ, Pan ZP, Liu SA, Sun J, Han M, Liu B, et al. The peptide encoded by a novel putative lncRNA HBVPTPAP inducing the apoptosis of hepatocellular carcinoma cells by modulating JAK/STAT signaling pathways. Virus Res. (2020) 287:198104. doi: 10.1016/J.VIRUSRES.2020.198104
9. Kusunoki H, Tanaka T, Kohno T, Kimura H, Hosoda K, Wakamatsu K, et al. NMR characterization of the interaction between Bcl-xL and the BH3-like motif of hepatitis B virus X protein. Biochem Biophys Res Commun. (2019) 518:445–50. doi: 10.1016/J.BBRC.2019.08.036
10. Bamdad T and Yari A. Hepatitis viruses. In: Encyclopedia of infection and immunity. Amsterdam: Elsevier. vol. 2. (2022). p. 113–22. doi: 10.1016/B978-0-12-818731-9.00078-1
11. Dagher J, Nassar M, and Faour WH. Harnessing AI for precision medicine and its applications in genomics, systems pharmacology, and drug discovery. Eur J Pharmacol. (2025) 1006:178183. doi: 10.1016/J.EJPHAR.2025.178183
12. Sen SK, Green ED, Hutter CM, Craven M, Ideker T, and Di Francesco V. Opportunities for basic, clinical, and bioethics research at the intersection of machine learning and genomics. Cell Genomics. (2024) 4:100466. doi: 10.1016/J.XGEN.2023.100466
13. Aruna AS, Remesh Babu KR, and Deepthi K. Autoencoder-based drug-virus association prediction with reliable negative sample selection: A case study with COVID-19. Biophys Chem. (2025) 322:107434. doi: 10.1016/J.BPC.2025.107434
14. Mulat M, Banicod RJS, Tabassum N, Javaid A, Kim TH, Kim YM, et al. Application of artificial intelligence in microbial drug discovery: Unlocking new frontiers in biotechnology. J Microbiol Methods. (2025) 237:107232. doi: 10.1016/J.MIMET.2025.107232
15. Jung Y, Grainger H, Yang S, Mondal S, Lukong KE, Conn K, et al. Catch me if you can: viral nucleic acids to host sensors. Front Immunol. (2025) 16:1632283/FULL. doi: 10.3389/FIMMU.2025.1632283/FULL
16. Jumper J, Evans R, Pritzel A, Green T, Figurnov M, Ronneberger O, et al. Highly accurate protein structure prediction with AlphaFold. Nature. (2021) 596:583–9. doi: 10.1038/s41586-021-03819-2
17. Tajwar R, Bradley DP, Ponzar NL, and Tavis JE. Predicted structure of the hepatitis B virus polymerase reveals an ancient conserved protein fold. Protein Sci. (2022) 31:e4421. doi: 10.1002/PRO.4421
18. Fasih F, Jabeen N, Hussain M, Habib A, Sharafat S, and Ul-Haq Z. Structural and antigenic variation in Hepatitis B virus oncogene, HBx. J Biomol Struct Dyn. (2025), 1–17. doi: 10.1080/07391102.2025.2517826
19. Hong X, Schneider WM, and Rice CM. Hepatitis B virus nucleocapsid assembly. J Mol Biol. (2025) 437:169182. doi: 10.1016/J.JMB.2025.169182
20. Sutradhar R and Dalal DC. Exploration of hepatitis B virus infection dynamics through an intracellular model. J Theor Biol. (2025) 611:112189. doi: 10.1016/J.JTBI.2025.112189
21. Karniadakis GE, Kevrekidis IG, Lu L, Perdikaris P, Wang S, and Yang L. Physics-informed machine learning. Nat Rev Phys. (2021) 3:422–40:2021. doi: 10.1038/s42254-021-00314-5
22. Raissi M, Perdikaris P, and Karniadakis GE. Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations. J Comput Phys. (2019) 378:686–707. doi: 10.1016/J.JCP.2018.10.045
23. Chen RTQ, Rubanova Y, Bettencourt J, and Duvenaud D. Neural ordinary differential equations. In: Advances in Neural Information Processing Systems (NeurIPS). (2018) 31:6571–83. doi: 10.5555/3327757.3327764
24. Wu C, Guo X, Li M, Shen J, Fu X, Xie Q, et al. DeepHBV: a deep learning model to predict hepatitis B virus (HBV) integration sites. BMC Ecol Evol. (2021) 21:138. doi: 10.1186/S12862-021-01869-8
25. Krause-Kyora B, Susat J, Key FM, Kühnert D, Bosse E, Immel A, et al. Neolithic and medieval virus genomes reveal complex evolution of hepatitis B. Elife. (2018) 7:e36666. doi: 10.7554/ELIFE.36666.001
26. Zlotnick A, Venkatakrishnan B, Tan Z, Lewellyn E, Turner W, and Francis S. Core protein: A pleiotropic keystone in the HBV lifecycle. Antiviral Res. (2015) 121:82–93. doi: 10.1016/J.ANTIVIRAL.2015.06.020
27. Gupta SP. Hepatitis viruses. In: Studies on hepatitis viruses. London: Academic Press (Elsevier). (2018). p. 15–33. doi: 10.1016/B978-0-12-813330-9.00002-8
28. Huang J, Zhao C, Zhang X, Zhao Q, Zhang Y, Chen L, et al. Hepatitis B virus pathogenesis relevant immunosignals uncovering amino acids utilization related risk factors guide artificial intelligence-based precision medicine. Front Pharmacol. (2022) 13:1079566/BIBTEX. doi: 10.3389/FPHAR.2022.1079566/BIBTEX
29. Zhang Y, Cao W, Wang S, Zhang L, Li X, Zhang Z, et al. Epigenetic modification of hepatitis B virus infection and related hepatocellular carcinoma. Virulence. (2024) 15:2421231. doi: 10.1080/21505594.2024.2421231
30. Shlesinger D, Hong KL, Shammas G, Page N, Sandu I, Agrafiotis A, et al. Single-cell immune repertoire sequencing of B and T cells in murine models of infection and autoimmunity. Genes Immun. (2022) 23:183–95. doi: 10.1038/S41435-022-00180-W;TECHMETA
31. Petkidis A, Andriasyan V, Murer L, Volle R, and Greber UF. A versatile automated pipeline for quantifying virus infectivity by label-free light microscopy and artificial intelligence. Nat Commun. (2024) 15:1–11. doi: 10.1038/S41467-024-49444-1;TECHMETA
32. Cuomo S, Di Cola VS, Giampaolo F, Rozza G, Raissi M, and Piccialli F. Scientific machine learning through physics–informed neural networks: where we are and what’s next. J Sci Computing. (2022) 92:3. 2022. doi: 10.1007/S10915-022-01939-Z
33. Avsec Ž, Agarwal V, Visentin D, Ledsam JR, Grabska-Barwinska A, Taylor KR, et al. Effective gene expression prediction from sequence by integrating long-range interactions. Nat Methods. (2021) 18:1196–203. doi: 10.1038/s41592-021-01252-x
34. Zhang H, Su X, Gu L, Tan M, Liu Y, Xu K, et al. Colchicine-mediated selective autophagic degradation of HBV core proteins inhibits HBV replication and HBV-related hepatocellular carcinoma progression. Cell Death Discov. (2024) 10:1–14. doi: 10.1038/S41420-024-02122-Z;SUBJMETA
35. Bell JT and Zhang X. The hepatitis B virus surface antigen: An evolved perfection and its unresolved mysteries. Virology. (2025) 608:110527. doi: 10.1016/J.VIROL.2025.110527
36. Yuan S, Xu J, Wang M, Huang J, Ma S, Liu Y, et al. HBV X protein induces degradation of UBXN7, a novel negative regulator of NF-κB signaling, to promote HBV replication. Cell Mol Gastroenterol Hepatol. (2023) 15:179–95. doi: 10.1016/J.JCMGH.2022.09.003
37. Zhao F, Deng Y, Liu W, and Li H. Spin repeats and human pathologies. In: Chromatin readers in health and disease. New York, NY: Academic Press (Elsevier). (2024). p. 257–77. doi: 10.1016/B978-0-12-823376-4.00013-6
38. Mobini S, Chizari M, Mafakher L, Rismani E, and Rismani E. Computational design of a novel VLP-based vaccine for hepatitis B virus. Front Immunol. (2020) 11:2074/BIBTEX. doi: 10.3389/FIMMU.2020.02074/BIBTEX
39. Yu Y, Kass MA, Zhang M, Youssef N, Freije CA, Brock KP, et al. Deep mutational scanning of hepatitis B virus reveals a mechanism for cis-preferential reverse transcription. Cell. (2024) 187:2735–2745.e12. doi: 10.1016/J.CELL.2024.04.008
40. Simon L. Advancing exposure science through artificial intelligence: Neural ordinary differential equations for predicting blood concentrations of volatile organic compounds. Ecotoxicol Environ Saf. (2025) 292:117928. doi: 10.1016/J.ECOENV.2025.117928
41. Galal AM, Haider Q, Hassan A, Arshad M, Alam MM, Al-Essa LA, et al. A besyian regularisation neural network approach for hepatitis B virus spread prediction and immune system therapy model. Sci Rep. (2024) 14:1–16. doi: 10.1038/S41598-024-75336-X;SUBJMETA
42. Huang D, Wu D, Wang P, Wang Y, Yuan W, Hu D, et al. End-of-treatment HBcrAg and HBsAb levels identify durable functional cure after Peg-IFN-based therapy in patients with CHB. J Hepatol. (2022) 77:42–54. doi: 10.1016/J.JHEP.2022.01.021
43. Kitagawa K, Kim KS, Iwamoto M, Hayashi S, Park H, Nishiyama T, et al. Multiscale modeling of HBV infection integrating intra- and intercellular viral propagation to analyze extracellular viral markers. PloS Comput Biol. (2024) 20:e1011238. doi: 10.1371/JOURNAL.PCBI.1011238
44. Erken R, Zaaijer HL, Willemse SB, Bakker E, Takkenberg BB, Reesink HW, et al. Hepatitis B core related antigen in relation to intrahepatic and circulating viral markers, before and after combination therapy. Ann Hepatol. (2021) 26:100540. doi: 10.1016/J.AOHEP.2021.100540
45. Dagnew M, Moges F, Tiruneh M, Million Y, Gelaw A, Adefris M, et al. Molecular diversity of hepatitis B virus among pregnant women in Amhara National Regional State, Ethiopia. PloS One. (2022) 17:e0276687. doi: 10.1371/JOURNAL.PONE.0276687
46. Murayama A, Yamada N, Toyama M, Aly HH, Nishitsuji H, Shimotohno K, et al. A synonymous mutation in the preS2 region enhances production of infectious hepatitis B virus. Virology. (2025) 612:110692. doi: 10.1016/J.VIROL.2025.110692
47. Trastoy R, Gonzalez-Alba JM, Soriano V, Rodriguez-Calviño JJ, de Mendoza C, Costa JJ, et al. The unexpected high prevalence of HBV subgenotype D4 in patients with chronic hepatitis B in Galicia, a northwestern Spanish region, reflects strong links with Latin America. J Clin Virology. (2022) 153:105195. doi: 10.1016/J.JCV.2022.105195
48. Wang M, Chen L, Dong MH, Li J, Zhu B, Yang Z, et al. Viral quasispecies quantitative analysis: a novel approach for appraising the immune tolerant phase of chronic hepatitis B virus infection. Emerg Microbes Infect. (2021) 10:842–51. doi: 10.1080/22221751.2021.1919033
49. Bouchard MJ, Puro RJ, Wang L, and Schneider RJ. Activation and inhibition of cellular calcium and tyrosine kinase signaling pathways identify targets of the HBx protein involved in hepatitis B virus replication. J Virol. (2003) 77:7713–9. doi: 10.1128/JVI.77.14.7713-7719.2003
50. Decorsière A, Mueller H, Van Breugel PC, Abdul F, Gerossier L, Beran RK, et al. Hepatitis B virus X protein identifies the Smc5/6 complex as a host restriction factor. Nature. (2016) 531:386–9. doi: 10.1038/NATURE17170
51. Kostyusheva A, Brezgin S, Glebe D, Kostyushev D, and Chulanov V. Host-cell interactions in HBV infection and pathogenesis: the emerging role of m6A modification. Emerg Microbes Infect. (2021) 10:2264–75. doi: 10.1080/22221751.2021.2006580
52. Lee CH, Huh J, Buckley PR, Jang M, Pinho MP, Fernandes RA, et al. A robust deep learning workflow to predict CD8 + T-cell epitopes. Genome Med. (2023) 15:70. doi: 10.1186/S13073-023-01225-Z
53. Kuhn R, Sandu I, Agrafiotis A, Hong KL, Shlesinger D, Neimeier D, et al. Clonally expanded virus-specific CD8 T cells acquire diverse transcriptional phenotypes during acute, chronic, and latent infections. Front Immunol. (2022) 13:782441/BIBTEX. doi: 10.3389/FIMMU.2022.782441/BIBTEX
54. Andriasyan V, Yakimovich A, Petkidis A, Georgi F, Witte R, Puntener D, et al. Microscopy deep learning predicts virus infections and reveals mechanics of lytic-infected cells. iScience. (2021) 24:102543. doi: 10.1016/J.ISCI.2021.102543
55. Meewan I, Panmanee J, Petchyam N, and Lertvilai P. HBCVTr: an end-to-end transformer with a deep neural network hybrid model for anti-HBV and HCV activity predictor from SMILES. Sci Rep. (2024) 14:1–24. doi: 10.1038/S41598-024-59933-4;SUBJMETA
56. Pavlova A, Fan Z, Lynch DL, and Gumbart JC. Machine learning of molecular dynamics simulations provides insights into the modulation of viral capsid assembly. J Chem Inf Model. (2025) 65:4844–53. doi: 10.1021/ACS.JCIM.5C00274
57. Sanchitra J, Debnath A, Singh AK, Jha AK, and Singh RK. Discovery of novel HBV core protein inhibitors by high throughput virtual screening. Sci Rep. (2025) 15:13054. doi: 10.1038/S41598-025-97242-6
58. Liu H, Rimal H, Lyu J, Shen L, Zhao Q, Chang J, et al. HBV capsid assembly modulators differentially modulate the assembly of wild-type and drug-resistant core protein chimeric nucleocapsids and empty capsids. PloS Pathog. (2025) 21:e1013391. doi: 10.1371/JOURNAL.PPAT.1013391
59. Wei S, Sasi C, Piepenbrock J, Huynen MA, and ‘t Hoen PAC. The use of knowledge graphs for drug repurposing: From classical machine learning algorithms to graph neural networks. Comput Biol Med. (2025) 196:110873. doi: 10.1016/J.COMPBIOMED.2025.110873
60. Baei B, Askari P, Askari FS, Kiani SJ, and Mohebbi A. Pharmacophore modeling and QSAR analysis of anti-HBV flavonols. PloS One. (2025) 20:e0316765. doi: 10.1371/JOURNAL.PONE.0316765
61. Odone A, Barbati C, Amadasi S, Schultz T, and Resnik DB. Artificial intelligence and infectious diseases: an evidence-driven conceptual framework for research, public health, and clinical practice. Lancet Infect Dis. (2025). doi: 10.1016/S1473-3099(25)00412-8
62. Rahimian M. Automated phage therapy in smart hospitals: a hypothesis for an AI, metagenomics, and IoT pipeline to treat resistant infections. Med Hypotheses. (2025) 203:111754. doi: 10.1016/J.MEHY.2025.111754
63. Gu X, Zheng M, Gao Y, Lin S, Zhang X, Chen C, et al. Overview of host-directed antiviral targets for future research and drug development. Acta Pharm Sin B. (2025) 15:1723–51. doi: 10.1016/J.APSB.2025.03.011
64. Banga JR and Villaverde AF. Mechanistic dynamic modelling of biological systems: The road ahead. Curr Opin Syst Biol. (2025) 42:100553. doi: 10.1016/J.COISB.2025.100553
65. Roquemen-Echeverri V and Mosquera-Lopez C. Recent advancements and applications of physics-informed machine learning in biomedical research. Curr Opin BioMed Eng. (2025) 35:100612. doi: 10.1016/J.COBME.2025.100612
66. Wang Z and Wu Z. Towards foundation model for chemical reactor modeling: Meta-learning with physics-informed adaptation. Chem Eng Res Design. (2025) 218:839–53. doi: 10.1016/J.CHERD.2025.05.015
67. Guo H, He K, Luo Y, and Chang Y. Physics-Informed Neural Networks for robust thermal comfort prediction: Overcoming data quality limitations through physiological constraints. Build Environ. (2025) 285:113588. doi: 10.1016/J.BUILDENV.2025.113588
68. Gezimati M and Singh G. Deep learning for multisource medical information processing. In: Data fusion techniques and applications for smart healthcare. Amsterdam: Elsevier/Academic Press. (2024). p. 45–76. doi: 10.1016/B978-0-44-313233-9.00009-6
69. Babar M, Qureshi B, and Koubaa A. Review on Federated Learning for digital transformation in healthcare through big data analytics. Future Generation Comput Systems. (2024) 160:14–28. doi: 10.1016/J.FUTURE.2024.05.046
70. Rybicka M, Woziwodzka A, Romanowski T, Sznarkowska A, Stalke P, Dręczewski M, et al. Host genetic background affects the course of infection and treatment response in patients with chronic hepatitis B. J Clin Virology. (2019) 120:1–5. doi: 10.1016/J.JCV.2019.09.002
71. Yang HI. Artificial intelligence in the prediction of progression and outcomes in viral hepatitis. In: Artificial intelligence, machine learning, and deep learning in precision medicine in liver diseases: concept, technology, application and perspectives. Amsterdam: Elsevier/Academic Press. (2023). p. 155–77. doi: 10.1016/B978-0-323-99136-0.00013-1
72. Talaminos-Barroso A, Reina-Tosina J, and Roa-Romero LM. Models based on cellular automata for the analysis of biomedical systems. In: Control applications for biomedical engineering systems. New York, NY: Elsevier/Academic Press. (2020). p. 405–45. doi: 10.1016/B978-0-12-817461-6.00014-7
73. Olsen M and Raunak M. Quantitative measurements of model credibility. In: Model engineering for simulation. Amsterdam: Elsevier/Academic Press. (2019). p. 163–87. doi: 10.1016/B978-0-12-813543-3.00008-1
74. Hallberg Szabadváry J and Löfström T. Beyond conformal predictors: Adaptive Conformal Inference with confidence predictors. Pattern Recognit. (2026) 170:111999. doi: 10.1016/J.PATCOG.2025.111999
75. von der Embse N, Winter S, Bonifay W, Kilgus S, Oddleifson C, Eklund K, et al. Reconciling discrepant universal screening data to improve decision-making: A Bayesian logistic regression approach. J Sch Psychol. (2025) 110:101461. doi: 10.1016/J.JSP.2025.101461
76. Wen J, Su A, Wang X, Xu H, Ma J, Chen K, et al. Virtual sample generation for small sample learning: A survey, recent developments and future prospects. Neurocomputing. (2025) 615:128934. doi: 10.1016/J.NEUCOM.2024.128934
77. Currie GM, Hawk KE, and Rohren EM. Generative artificial intelligence biases, limitations and risks in nuclear medicine: an argument for appropriate use framework and recommendations. Semin Nucl Med. (2025) 55:423–36. doi: 10.1053/J.SEMNUCLMED.2024.05.005
78. Khalid MNA, Ullah A, Numan M, and Majid A. Words shaping worlds: A comprehensive exploration of text-driven image and video generation with generative adversarial networks. Neurocomputing. (2025) 632:129767. doi: 10.1016/J.NEUCOM.2025.129767
79. Bradley JM, Hutchings M, Arden MA, O’Cathain A, Maguire C, and Wildman MJ. A RCT to explore the effectiveness of supporting adherence to nebuliser medication in adults with cystic fibrosis: fidelity assessment of study interventions. BMC Pulm Med. (2024) 24:1–12. doi: 10.1186/S12890-024-02923-Z/FIGURES/3
80. Nih K. PROPOSED BIOSECURITY OVERSIGHT FRAMEWORK FOR THE FUTURE OF SCIENCE. Bethesda, MD: National Institutes of Health (National Science Advisory Board for Biosecurity) (2023).
81. United States Government. United states government policy for institutional oversight of life sciences dual use research of concern key dates. Washington, DC: United States Government (2014).
Keywords: hepatitis B virus, artificial intelligence, machine learning, cccDNA, viral replication
Citation: Hu Y and Jiang W (2025) Artificial intelligence for mechanistic understanding of hepatitis B virus. Front. Virol. 5:1729171. doi: 10.3389/fviro.2025.1729171
Received: 21 October 2025; Accepted: 07 November 2025; Revised: 05 November 2025;
Published: 27 November 2025.
Edited by:
Humberto J. Debat, Instituto Nacional de Tecnología Agropecuaria, ArgentinaReviewed by:
Viswanath Buddolla, Dr. Buddolla’s Institute of Life Sciences, IndiaCopyright © 2025 Hu and Jiang. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
*Correspondence: Wei Jiang, amlhbmd3ZWk4NjZAcWR1LmVkdS5jbg==