The Role of Artificial Intelligence and Machine Learning in Predictive Virology: Forecasting,Tracking,and Combating Viral Threats

Abstract

The escalating threat of viral pandemics, dramatically illustrated by the COVID-19 crisis, has exposed the critical shortcomings of conventional reactive virology in addressing rapidly evolving pathogens. This review introduces predictive virology (PV) as an artificial intelligence (AI)-driven discipline within broader epidemic intelligence and public health surveillance that uses advanced computational tools to forecast viral threats and accelerate countermeasure design. The current review systematically examines how AI-driven approaches (e.g., machine learning and deep learning) are reshaping virology by integrating vast genomic datasets, multimodal surveillance signals, and advanced computational models to anticipate viral emergence and evolution before widespread transmission occurs. Core pillars of PV discussed include zero-shot mutational fitness and antigenic escape prediction using large protein language models; multimodal early-warning systems that fuse wastewater monitoring, digital epidemiology, mobility data, and social media; neural differential equation-based transmission modeling; generative AI for de novo design of broad-spectrum antivirals and vaccines; and ecological risk assessment of zoonotic spillovers. In retrospective benchmarks against deep mutational scanning experiments and real-world epidemiological outcomes (SARS-CoV-2 variants, influenza, and other outbreaks), several AI-powered tools have demonstrated performance comparable to or exceeding traditional methods, although prospective validation at scale remains limited. Despite remarkable progress, significant challenges persist, including data bias, overfitting to historical patterns, lack of prospective validation, and limited generalizability across settings. In addition, there are concerns about mechanistic interpretability, equitable global data integration, and responsible deployment. This review also critically addresses the ethical, governance, and equity implications of deploying predictive capabilities at a global scale. By consolidating cutting-edge AI methodologies with virological insights and acknowledging current limitations, this work provides a comprehensive framework for transitioning virology from a reactive to a truly predictive discipline, ultimately strengthening global health security and pandemic preparedness.

Keywords

artificial intelligence machine learning predictive virology viral evolution epidemic forecasting antiviral design

Introduction

Viruses can parasitize all living organisms and microorganisms, including humans, animals, plants, bacteria, fungi, and algae. Virus infections have been linked with a variety of diseases in humans. For example, the Spanish flu, the 1918 influenza pandemic, resulted in the death of approximately 100 million people (Aassve et al., 2021). Since its discovery in 1983, approximately 40.8 million people were living with HIV at the end of 2024, and 44.1 million people have died from AIDS-related illnesses (UNESCO, 2021). In addition, Ebola virus disease is among the most lethal viral infections, with case fatality rates varying from 25% to 90% across outbreaks (average ∼50%) and the Zaire ebolavirus species reaching up to 90% in some settings. The virus frequently attacks West Africans, with the largest outbreaks having primarily occurred in Central and West Africa due to ecological and epidemiological factors, with the largest outbreak reported in 2014–2016, with 28,600 reported cases ((CDC) CfDCaP, 2024). Recently, SARS-CoV-2, the causative agent of COVID-19, caused the global pandemic, with 778,900,250 laboratory-confirmed cases and 7,103,247 deaths (WHO, 2023). These viral epidemics and pandemics occurred with limited advance warning, highlighting the shortcomings of traditional reactive virology (Bedford et al., 2020). The rapid evolution of viruses, particularly RNA viruses, poses a challenge to the development and approval of traditional vaccines, which require years. In addition, anthropogenic ecosystem disruption, urbanization, and climate change accelerate viral spillover (Carlson et al., 2022). Another important factor that mediates rapid virus dissemination is global air travel networks that can seed a pandemic virus on several continents within hours (Findlater and Bogoch, 2018; Mastala et al., 2026).

Following the clinical recognition of newly emerged viruses, genomic characterization has taken weeks, the development of diagnostic assays has been achieved in weeks to months, and the approval of vaccines or therapeutics has taken years. In addition, applying countermeasures imposes enormous human and economic costs. The global economic losses from the COVID-19 pandemic have been estimated at approximately US$14 trillion in lost output and health reductions, with broader projections indicating cumulative global GDP losses on the order of tens of trillions of dollars (Cutler and Summers, 2020; Nie et al., 2025). Currently, three technological revolutions have been developed to alleviate the impacts of virus epidemics and pandemics on human health and the economy: (i) advanced next-generation sequencing (NGS) and third-generation single-molecule long-read technologies used for rapid virus discovery and real-time surveillance of emerging variants. These technologies are cost-effective in high-throughput settings and can generate complete viral genomes in a few hours (often <US$100 per sample for reagent costs in optimized labs; Shi et al., 2024). (ii) Open data-sharing platforms such as GISAID, GenBank, and the INSDC network. These databases secure more than 25 million viral genomes with globally accessible and real-time resources (Feng et al., 2024). (iii) Rapid advancements in artificial intelligence (AI) tools, particularly deep neural networks, have occurred. AI and its branches (machine learning [ML] and deep learning [DL]) can process billions to hundreds of billions of parameters and predict signals from noisy, high-dimensional datasets at a scale that humans or classical statistical methods cannot achieve (Hadfield et al., 2018; Lin et al., 2023; Sinno et al., 2025; Fig. 1). Collectively, these three advancements have led to the birth of predictive virology (PV), the focus of this review. PV positions AI tools (ML and DL) as its foundational engine, enabling proactive forecasting of viral threats and preemptive countermeasure design rather than relying solely on reactive responses. Although the term PV is not yet widely indexed in bibliographic databases, this review systematically assembles its five core pillars, as detailed in the “Methods” section. The first pillar focuses on genomic surveillance and zero-shot prediction of mutational fitness and antigenic evolution using protein language models (pLMs; the “Genomic Surveillance in the Era of Big Data: AI-Driven Mutation Prediction and Variant 2 Emergence” section). The second pillar encompasses multimodal real-time epidemic intelligence and early-warning systems integrating diverse data streams such as wastewater monitoring, digital epidemiology, mobility data, and social media (the “Real-Time Epidemic Intelligence: ML for Early Detection and Nowcasting” section). The third pillar addresses high-resolution transmission and intervention modeling with neural differential equations and hybrid approaches (the “Transmission Dynamics and Forecast Modeling: Beyond Classical Compartmental Models” section). The fourth covers AI-accelerated design and optimization of broad-spectrum antivirals, vaccines, and monoclonal antibodies (the “Antiviral Discovery and Vaccine Design Accelerated by DL” section). Last, the fifth pillar involves ecological and zoonotic spillover risk assessment at the animal–human–environment interface (integrated in the “The Emerging Discipline of PV: Definitions, Scope, and Imperatives” section and Fig. 2). Each pillar is benchmarked against experimental and real-world data, with cross-cutting challenges discussed in later sections.

The main task of this proactive discipline is to seek not only to describe epidemics that have already occurred but also to forecast future epidemics. In addition, AI models show promise in anticipating where the next variant may emerge, how transmissible and immune-evasive it could be, and in providing forecasts that could help guide countermeasures months in advance, although widespread operational use at this scale remains aspirational (Carlson et al., 2021). However, prospective, real-world validation at scale is still lacking, and significant practical limitations remain, including data bias, generalizability to novel viruses, and the absence of operational deployment frameworks.

The high-throughput genomic data obtained by NGS represent a challenge to traditional tools such as BLAST and HMMER. These tools have become unsuitable for detecting highly divergent viral genomes where much of the virome remains unexplored (the so-called “dark matter” of viral sequences; Blanco-Melo et al., 2024). AI tools such as ML models and DL architectures have been developed to overcome the limitations of traditional methods. ML-based methods using k-mer frequency distributions as features have the ability to differentiate between viral and nonviral genomic sequences (Sironi and Kaderali, 2021). While convolutional neural networks and recurrent neural networks remain useful, transformer-based models and large pLMs now represent the state of the art for capturing long-range dependencies and hierarchical features in viral sequences. Therefore, these AI tools allow sensitive detection, classification, and host prediction of novel viruses. In addition, AlphaFold and ESMFold, structure-based models, can annotate proteins through conserved three-dimensional (3D) folds (Sironi and Kaderali, 2021; Bowyer et al., 2025). AI tools can be employed in public health because of their ability to integrate a variety of real-time data, health records, surveillance data, environmental factors, and even social media. The processing of such large amounts of data resources enables forecasting of outbreaks, prediction of transmission patterns, and identification of hotspots that offer proactive early warning systems, which is especially valuable in remote regions (Pagsuyoin et al., 2025).

Another interesting role of AI in public health is in drug discovery by considerably shortening the traditional slow and costly pipeline timelines from years to months. AI-driven approaches perform rapid screening and predict drug–target and protein–ligand interactions via graph neural networks (GNNs) and transformers, often achieving strong performance on benchmarks, although real-world validation gaps remain (Kupperman et al., 2022; Chen et al., 2025). AI models can also be extended to personalized patient care and generate targeted therapeutics for neglected and rare diseases (Baig et al., 2024). While PV shares substantial methodological overlap with epidemic intelligence and digital epidemiology, its distinctive strength lies in leveraging deep molecular insights from pLMs and generative AI to move beyond early detection toward preemptive countermeasure development. The current review aims to critically evaluate the development of PV as a proactive discipline capable of forecasting the emergence of novel viruses and variants, viral evolution, and epidemic potential before widespread transmission occurs. While AI tools offer powerful new capabilities, they are not a panacea. Several important limitations in mechanistic interpretability, data bias, generalizability to novel viruses, and real-world validation persist. This review systematically examines the core AI-driven pillars of PV: (i) zero-shot mutational effect prediction (i.e., for novel viruses or variants without training on labeled data from those targets) via evolutionary-scale pLMs; (ii) multimodal early-warning systems that integrate wastewater, digital exhaust, and mobility networks; (iii) continuous-time transmission modeling via neural ordinary differential equations and graph transformers; (iv) generative design of broad-spectrum antivirals and vaccines along with quantitative forecasting of antigenic drift and immune escape; and (v) health risk stratification of zoonotic reservoirs. For each pillar, state-of-the-art performance is benchmarked against gold-standard deep mutational scanning experiments and real epidemiological outcomes from the SARS-CoV-2 pandemic and subsequent outbreaks (2020–2025).

Methods

This review was conducted as a narrative synthesis with systematic elements to ensure comprehensive coverage of the rapidly evolving field of PV. The aim was to identify key conceptual developments, state-of-the-art AI tool applications, benchmarking studies, and critical discussions relevant to forecasting, tracking, and combating viral threats.

Data sources

Literature was searched in the following major databases: PubMed/MEDLINE, Scopus, Web of Science, and Google Scholar. Gray literature, including reports from the World Health Organization (WHO) and Centers for Disease Control and Prevention (CDC), was also consulted where relevant.

Search strategy and terms

The core search strategy combined terms related to AI, ML, and DL with virology and predictive/epidemiological concepts. Because PV is an emerging conceptual framework rather than a widely indexed keyword, a search limited to this exact phrase would retrieve only a handful of publications and would fail to capture the full breadth of relevant AI-driven research. Therefore, a component-based (pillar-driven) search strategy was adopted. The five core technological pillars of PV (as outlined in the Introduction) guided the development of specific technical search strings. These included combinations such as (“protein language model” AND “viral fitness”), (“wastewater” AND “deep learning” AND “early warning” OR “nowcasting”), (“neural ODE” AND “epidemic forecasting”), (“generative AI” OR “generative model” AND “antiviral design”), and (“zoonotic spillover” OR “spillover risk” AND “machine learning”). These pillar-specific terms were combined with general AI/ML terms (“artificial intelligence,” “machine learning,” and “deep learning”) and virological/epidemiological outcome terms (e.g., “viral evolution,” “antigenic drift,” “immune escape,” “transmission modeling,” and “vaccine design”). The phrase “predictive virology” was included as an umbrella term to capture any articles that explicitly used it, but it was not relied upon as the primary filter. Filters were applied for English-language publications (Fig. 3).

Inclusion and exclusion criteria

Articles were included if they described AI/ML models including pLMs, multimodal systems, neural differential equations, generative models, and similar approaches, applied to areas such as viral mutation prediction, antigenic evolution, epidemic nowcasting or forecasting, zoonotic risk assessment, or AI-accelerated antiviral and vaccine design. In addition, the studies had to provide benchmarking against experimental data (e.g., deep mutational scanning) or real-world epidemiological outcomes, and they were required to offer conceptual, methodological, or critical insights into PV or related ethical and equity issues. Articles were excluded if they focused solely on classical (non-AI) virological or epidemiological methods without meaningful AI integration, were purely clinical case reports or therapeutic trials without any computational modeling component, or lacked sufficient methodological detail and were not peer-reviewed (with the exception of high-impact preprints or official reports; Fig. 3).

The Emerging Discipline of PV: Definitions, Scope, and Imperatives

PV is defined as an emerging disciplinary field that integrates advanced AI tools, particularly pLMs, generative models, neural differential equations, and multimodal fusion systems, with virological, genomic, ecological, and epidemiological data to proactively forecast viral emergence, evolutionary trajectories (including mutational fitness, antigenic escape, and immune escape), transmission dynamics, and zoonotic spillover risk, while accelerating the de novo design of broad-spectrum countermeasures such as antivirals and vaccines (Spreco et al., 2022; Pagsuyoin et al., 2025). PV builds directly upon and extends established frameworks such as epidemic intelligence (which encompasses traditional surveillance, digital epidemiology, and infodemiology using open-source, social media, and syndromic data for outbreak detection) and broader public health surveillance (Gao, 2022; Kaur and Butt, 2025). While there is considerable overlap, particularly in real-time early warning and nowcasting, PV is distinguished by its strong emphasis on molecular-scale predictive modeling of viral evolution and preemptive therapeutic/vaccine design using evolutionary-scale pLMs and generative AI (Liu et al., 2025; UNAIDS, 2025). These capabilities enable foresight into variant fitness and countermeasure optimization months ahead of empirical data, shifting virology from purely reactive response toward genuine molecular preemption. It could reasonably be viewed as a specialized, virology-centric pillar within the broader umbrella of epidemic intelligence, with the unique addition of AI-accelerated control measure design (Suri and Dakshanamurthy, 2022; Kaur and Butt, 2025; Thomason, 2026). In this framework, AI tools are not merely supportive tools but the central integrative engine that unifies genomic prediction, multimodal surveillance, transmission modeling, countermeasure design, and zoonotic risk assessment into a cohesive predictive discipline.

The scope of PV is categorized into five interrelated pillars that build upon and enhance traditional public health surveillance and epidemic intelligence: (i) genomic surveillance and zero-shot prediction of mutation fitness and antigenic evolution using pLMs; (ii) multimodal real-time epidemic intelligence and early-warning systems (integrating wastewater, digital exhaust, mobility, and social media data); (iii) high-resolution transmission and intervention modeling; (iv) AI-accelerated design and quantitative optimization of control measures (broad-spectrum antivirals, vaccines, and monoclonal antibodies); and (v) ecological and zoonotic spillover risk assessment at the animal–human–environment interface. Pillars (i), (ii), (iii), and (v) significantly overlap with and strengthen existing public health surveillance systems. However, Pillar (iv) represents a distinct proactive capability that goes beyond detection and forecasting to enable preemptive molecular interventions. This integration of deep virological prediction with actionable countermeasure design is a core differentiator of PV.

Each approach produces data to feed the others. For example, predicted mutation fitness models inform variant-of-concern designation algorithms. Consequently, early warning signals have begun to trigger adaptive sequencing strategies, and transmission forecasts are being refined to guide preemptive vaccine strain selection. Finally, spillover risk maps are utilized to direct targeted field sampling (Keeling and Rohani, 2008; Grange et al., 2021; Ebrahimi and Ghaemi, 2026). As of late 2025, these approaches are being actively developed and conducted for influenza, SARS-CoV-2, Respiratory Syncytial Virus (RSV), and selected arboviruses. In several retrospective studies, AI-based models have predicted influenza antigenic drift with accuracy comparable to or exceeding traditional laboratory-based methods, although prospective validation remains limited (Ming et al., 2023; Shah et al., 2025). In well-sampled urban settings with robust wastewater and digital surveillance infrastructure, cryptic transmission clusters have been detected up to 3 weeks before hospital surges; however, this performance depends heavily on data quality and local context (Rehms et al., 2024; Sudhakar et al., 2024). In addition, broadly neutralizing pan-family inhibitors can now be designed in silico, and unknown animal viruses can be ranked by spillover potential years before human transmission (Fig. 2; Grange et al., 2021; Morbey et al., 2023).

Despite these promising capabilities, predictive models in PV face important limitations that must be acknowledged. Data bias is pervasive in that most training datasets overrepresent high-income countries and clinically detected variants, which can lead to models that underperform in undersampled regions or for neglected viruses (Popov et al., 2023; Flores et al., 2024). Overfitting to historical patterns is a persistent risk, especially when models are evaluated only on the same data used for training (Mulomba et al., 2025). Furthermore, results obtained in one geographic or epidemiological setting often fail to generalize to others due to differences in population structure, health care seeking behavior, viral lineages, and surveillance intensity. Prospective validation (i.e., testing model predictions against future unseen data) remains rare, and real-world operational deployment is still in early stages (Popov et al., 2023). These limitations are further discussed in the “Ethical, governance, and equity challenges in PV” and “Conclusions” sections.

The imperatives driving the importance of PV can be summarized in the following points: (i) predicting viral emergence and improving pandemic preparedness. At least 10,000 virus species are estimated to have the potential to infect humans, most of which currently circulate silently in wild mammals (Carlson et al., 2019). Climate change and land-use alterations are expected to facilitate zoonotic spillover, with more than 4000 mammalian viruses predicted to shift their geographic range by 2070 and significantly increase overlap with human populations (Bartlow et al., 2019; Carlson et al., 2022; Holmes et al., 2024). (ii) Predicting viral evolution and immune evasion. PV employs ML and DL models to forecast the kinetics of viral evolution and to identify key mutations responsible for drug resistance and immune escape, thereby enabling timely redesign of vaccines and antiviral therapies (Mello and Wang, 2020; Blassel et al., 2021). (iii) Accelerating the development of novel countermeasures. Through the integration of AI, PV can significantly shorten traditional drug and vaccine development timelines by rapidly identifying and prioritizing promising therapeutic targets (Shafi and Parwani, 2023; Rajput et al., 2024; Pagsuyoin et al., 2025). (iv) Systematic classification and characterization of the global virosphere. Only a small fraction of the world’s viruses have been discovered and characterized. Through metagenomics and computational methods, PV supports the systematic discovery and classification of the vast, uncategorized virome (Holmes et al., 2024). (v) Efficient handling and analysis of large-scale multimodal data. Modern sequencing technologies generate enormous volumes of genomic, mobility, clinical, and environmental data that exceed the capacity of traditional analytical tools. PV can analyze and interpret this “big data” effectively (Holmes et al., 2024; Pagsuyoin et al., 2025).

Integrating the five pillars: Challenges and barriers

The five pillars of PV described above are not independent; their value emerges from integration. Predicted mutational fitness [Pillar (i)] can flag emerging variants for multimodal surveillance [Pillar (ii)]. Early warning signals can trigger adaptive sequencing and refine transmission models [Pillar (iii)]. Spillover risk maps [Pillar (v)] can guide targeted sampling that feeds back into genomic databases. In addition, predictions from Pillars (i), (ii), (iii), and (v) can inform the design of broad-spectrum countermeasures [Pillar (iv)], creating a pathway from early detection to preemptive intervention. In an ideal PV system, these loops would operate in near real time, shortening the detection-to-response timeline to weeks or days (Fig. 2). However, several barriers currently prevent such seamless integration. First, data interoperability remains a major hurdle: genomic sequences, wastewater RNA concentrations, mobility data, and clinical case reports use different formats, temporal resolutions, and quality standards, making automated fusion difficult (Hassan Nishan, 2026). Second, institutional silos separate academic virology, public health surveillance, clinical medicine, and environmental monitoring; data sharing across these domains remains ad hoc and often delayed (FAO FaAOotUN, 2025; Matson et al., 2024). Third, real-time data access is constrained by privacy regulations e.g., General Data Protection Regulation (GDPR) and Health Insurance Portability and Accountability Act (HIPAA), commercial confidentiality (mobility data), and national security concerns (pathogen genomics; Ali Maher et al., 2026; Bakas et al., 2026).

Fourth, computational integration is nontrivial: running a pLM alongside a neural ODE (i.e., a neural network that parameterizes continuous-time dynamics via ordinary differential equations) and a multimodal transformer in a unified pipeline requires substantial engineering and low-latency infrastructure that is rarely available outside a few high-resource centers (Thompson et al., 2022). Fifth, feedback validation loops are almost entirely missing in that predictions are rarely used to prospectively guide interventions, and even when they are, outcomes are not systematically fed back to improve the models (Yang et al., 2022). Partial integration exists in platforms such as Nextstrain (genomic epidemiology), GISAID (data sharing), and WHO Epidemic Intelligence from Open Sources (EIOS; open-source intelligence), but these operate largely in parallel rather than as a unified PV system (Hadfield et al., 2018). Overcoming these barriers will require not only technical solutions (standardized application programming interfaces, federated learning, and privacy-preserving computation) but also governance reforms (incentives for data sharing and international agreements on real-time pathogen data exchange), issues further discussed in the “Ethical, Governance, and Equity Challenges in PV” and “Future Directions” sections. Thus, while PV draws heavily upon existing frameworks such as epidemic intelligence and digital epidemiology, its defining feature is the central, synergistic role of AI and ML—particularly evolutionary-scale pLMs and generative approaches—in enabling true molecular-level foresight and preemptive countermeasure design. The subsequent sections examine each of the five pillars of PV in turn, demonstrating in each case how AI/ML provides the predictive capability that distinguishes PV from traditional virology.

Genomic Surveillance in the Era of Big Data: AI-Driven Mutation Prediction and Variant Emergence

The first pillar of PV leverages AI-driven genomic surveillance and zero-shot mutational fitness and antigenic evolution prediction using large pLMs. Currently, publicly accessible databases such as the GISAID, GenBank, and INSDC networks contain millions of viral genomes of SARS-CoV-2 (18.2 million), influenza (4.1 million), and RSV (1.8 million). In addition, rapidly growing genome collections of enteroviruses, dengue virus, monkeypox virus, and avian influenza A virus have been reported. This large genome repository constitutes the largest dataset ever collected for any group of organisms (Hadfield et al., 2018; Sinno et al., 2025). The classical bioinformatic tools used for sequence and phylogenetic analysis and operated by human experts can handle only hundreds to a few thousand genomic sequences. In addition, tree construction now requires weeks on supercomputers, mutation annotation is prone to error, and no individual or team can meaningfully analyze tens of thousands of new genomes per day. DL models such as pLMs are feasible tools for analyzing large amounts of data. pLMs with 650 million to 15 billion parameters are pretrained unsupervised on 200–500 million diverse protein sequences and can serve as universal mutational effect predictors. They operate in milliseconds per variant and achieve Spearman correlations of 0.85–0.92 against gold-standard deep mutational scanning libraries for spike glycoprotein, hemagglutinin (HA), neuraminidase (NA), and viral polymerases—substantially higher than alignment-based methods such as EVmutation (typically ρ ∼0.60–0.70; Hopf et al., 2017; Liu et al., 2020; Brandes et al., 2023). However, these high correlations can mask important limitations. pLMs often underpredict the fitness effects of rare beneficial mutations that are poorly represented in their pretraining corpora and exhibit biases against viral proteins due to their sparse representation in training datasets (Sebastianelli et al., 2024; Gibson et al., 2026). Moreover, extreme sequence homology among rapidly evolving viruses can lead to “data leakage” in standard random validation splits, yielding inflated performance metrics that fail to translate into real-world biosurveillance utility (Gibson et al., 2026). pLMs also cannot reliably handle insertions or deletions, a challenge exacerbated by altered protein lengths and limited annotated datasets (Gracia Carmona et al., 2026), and false-positive predictions of deleterious mutations as neutral occur, particularly for viral proteins with few sequence homologs. As noted by Sinno et al., AI-driven approaches in virology face challenges including dataset bias, elevated false discovery rates, and limited explainability, issues that are particularly relevant when models are applied to undersampled viral families or novel variants (Sironi and Kaderali, 2021). These limitations are further discussed in the “Ethical, Governance, and Equity Challenges in PV” section.

DL models have effectively shortened the previously required years of wet-lab experimentation (Table 1). The most transformative application lies in the prediction of prospective antigenic evolution. Early substitution-based models (e.g., dN/dS ratios and site-specific selection coefficients) address past evolution. Modern multimodal neural networks act by combining several models, such as the primary sequence, AlphaFold-predicted structure, electrostatic surface properties, and historical serological maps, to predict antigenic distance before new virus variants dominate (Hardy-Werbin et al., 2023; Islam et al., 2025). The challenge of influenza prediction illustrates the need for AI-driven genomic surveillance. Predictive evolutionary modeling focuses on the highly mutable HA protein of influenza, which is the primary target for the selection of vaccine strains. Compared with traditional expert-led analysis and heuristic methods, new computational tools demonstrate superior performance in processing millions of influenza genomic sequences. The beth-1 model has outstanding performance in anticipating viral evolution and selecting vaccine strains for influenza A subtypes (Loucera et al., 2020).

Table 1.

Summary of Key Artificial Intelligence, Machine Learning, and Deep Learning Approaches in Predictive Virology

Model/Framework	AI technique	Target virus/System	Application/Task	Key outcome/Result	Reference
CoVFit	Protein language model (ESM-2 adaptation)	SARS-CoV-2	Viral fitness landscape prediction	Predicts variant fitness based solely on spike protein sequences, providing a tool for monitoring emerging variants	(The Genotype to Phenotype Japan C, 2025)
ViTax	DL (adaptive hierarchical)	General (viral sequences)	Viral taxonomic classification	Provides precise, adaptive hierarchical classification of viral sequences, essential for novel virus discovery and functional assessment	(He et al., 2024)
ConCreT	2D convolutional neural network (CNN)	Phylum Cressdnaviricota	Viral taxonomic classification	Achieves high accuracy (>98%) in classifying viruses within a specific phylum, demonstrating the power of CNNs for sequence-based taxonomy	(Gomes and Zerbini, 2023)
E2VD (evolution-driven virus variation driver prediction)	DL (evolution-driven framework)	SARS-CoV-2, influenza, Zika, HIV	Prediction of virus variation drivers and evolutionary fitness	Unified, structure-free, and interpretable approach that significantly outperforms state-of-the-art methods in predicting beneficial mutations	(Nikparvar et al., 2021)
ViralForesight/GenPreMut	DL (generative framework), protein language models (pLMs)	SARS-CoV-2	Generative prediction of real-world prevalent mutations using in silico virus evolution	Accurately predicts future prevalent mutations by simulating within-host to population-level selective pressures	(Livieratos et al., 2025)
MLAEP (machine learning-guided antigenic evolution prediction)	ML (structure modeling, multitask learning, genetic algorithms)	SARS-CoV-2	Forecasting antigenic evolution and immune escape	Combines multiple computational techniques to predict the evolution of the spike protein’s receptor-binding domain (RBD).	(Han et al., 2023)
VIDHOP	DL	General (viral genomes)	Viral host prediction (based on genome sequence)	Fast and accurate host prediction using only viral genome sequence, even with short sequence fragments (100–400 bp)	(Moein et al., 2021)
HostNet	DL (transformer-CNN-BiGRU)	General (viral genomes)	Viral host prediction (sequence representation)	Improved sequence representation for virus–host prediction, overcoming challenges in sequence-based classification	(Mock et al., 2021)
Deep learning model	DL (general)	HIV	Real-time outbreak detection (using genetic data)	Effective strategy for real-time outbreak detection using genomic data, identifying subsets of sequences from active outbreaks	(Lai et al., 2023)
DeepHVI	Multimodal DL Protein language models (ESM-2, LucaProt)	Human-virus protein–protein interactions, specifically SARS-CoV-2-human interactions	Human-virus protein–protein interaction (HV-PPI) prediction	Integrates protein sequence data to predict critical interactions, advancing viral research and drug target identification	(Watson et al., 2021)
Multistage deep learning framework (model 2)	Long short-term memory networks, transformer architecture, self-attention mechanisms	COVID-19	Spatio-temporal epidemic forecasting using mobility data	Model 2 (LSTM + transformer with mobility data) outperforms baseline approaches; achieves lower error rates across multiple metrics; inclusion of mobility data improves predictive accuracy, especially for longer-term forecasts	(Jiao et al., 2025)
Multilayer perceptron (MLP), bidirectional recurrent neural network (BRNN), and convolutional neural network (CNN)	DL techniques (MLP, BRNN, CNN)	HIV-1	Drug resistance prediction for 18 antiretroviral therapy (ART) drugs using viral genotype–phenotype data	High classification performance is dependent on drug resistance mutations (DRMs); CNN was the best-performing architecture; model interpretability revealed that the models heavily weighted features that are not known DRM locations, indicating utility for addressing causal relationships	(Stone et al., 2026)
EVEscape	DL combined with other models	SARS-CoV-2 virus, with applicability extended to diverse virus types	Predicting potential viral escape mutations. Analyzing virus mutation. Aiding vaccine design and public health measures	Prediction accuracy exceeding 85%; capable of predicting mutation trajectories based on historical viral sequence data	(Lin et al., 2022)
Gaussian process (GP)-based spatial covariance (SCV)	ML, Gaussian process (GP) modeling, spatial covariance (SCV), early warning anomaly detection (EWAD)	SARS-CoV-2	Early-warning pandemic tracking	EWAD can anticipate changes in the pattern of spread and pathology weeks in advance, identifying signatures destined to become Variant of Concern (VOCs); GP-based SCV relationships, in conjunction with co-occurrence analysis, provide a powerful tool for genomic surveillance	(Lou et al., 2024)
MFPAD framework (multifeature-based prediction of antigenic distance)	Multitask low-rank matrix completion, XGBoost regression model	Influenza H3N2	Predicting antigenic distances between viruses	The MFPAD framework for H3N2 influenza prediction shows superior performance with low prediction error and high accuracy in predicting antigenic variants. It successfully identifies major antigenic clusters, aligns with serological data, and outperforms other models	(Li et al., 2024)
ML model framework	Ensemble model (GBM + LSTM + SVM)	Dengue	Dengue incidence prediction	Ensemble model for forecasting dengue outbreaks using environmental and geospatial data	(Serrano et al., 2024)
XAI model	Explainable AI (SHAP-integrated ML)	COVID-19	Prognostic severity prediction	Explainable AI for prognostic severity prediction from clinical biomarkers	(Chadaga et al., 2024)
VAE-ML model	Variational autoencoder (VAE) and ML	COVID-19	Therapeutic efficacy prediction	Predicts the activity of COVID-19 therapeutics and vaccines against emerging variants	(McMahan et al., 2021)

AI, artificial intelligence; DL, deep learning; ML, machine learning.

Deep mutational scanning of the HA protein provides valuable data on the effects of single amino acid mutations on viral fitness, which can be used to predict the evolutionary fates of H3N2 lineages in nature (Li et al., 2020). Additionally, the integration of evolutionary data into mechanistic epidemiological models has allowed for accurate prediction of seasonal H3N2 incidence in advance of the season (Du et al., 2017). ML is also being applied to predict the seasonal antigenic properties of H3N2, accurately anticipating hemagglutination inhibition assay results to assist in vaccine strain selection (Shah et al., 2025). Similarly, the large amount of genomic data on SARS-CoV-2 needs the power of ML and DL to analyze and predict the antigenic impact of virus variants, particularly with rapid virus evolution (Table 1). Zero-shot models were tested for their ability to predict the antigenic impacts of omicron variants of SARS-CoV-2. These models were first trained on pre-Omicron data and successfully predicted the antigenic effects of the variants BA.2.86, XBB.1.16, and JN.1 lineages months before neutralization data became available. These models are able to precisely determine amino acid substitutions, R346T, L368I, and F486P substitutions, that define 2024–2025 immune escape (Table 1) (Naseer et al., 2022; Cheng et al., 2024). Collectively, these AI-driven genomic surveillance approaches constitute the first pillar of PV, transforming retrospective sequence analysis into prospective forecasting of variant fitness and antigenic evolution.

Real-Time Epidemic Intelligence: ML for Early Detection and Nowcasting

The second pillar of PV integrates multimodal AI systems for real-time epidemic intelligence and early-warning systems that fuse diverse data streams such as wastewater monitoring, digital epidemiology, mobility data, and social media. Early during an epidemic, identifying the causative agent is crucial for controlling virus spread, applying appropriate control measures, and developing drug or vaccine candidates. The classical approaches adopted by public health rely on physician-initiated reporting of notifiable diseases, laboratory confirmation, and weekly aggregation. Such approaches usually introduce irreducible delays of 7–21 days even in high-income settings and far longer in low-resource regions. Considerable efforts have been made to estimate key epidemiological parameters of SARS-CoV-2, such as the reproduction number (R₀), which has been determined and publicized (Abbott and Thompson, 2020; Tonks et al., 2024). The importance of real-time reporting (i.e., nowcasting) and forecasting was demonstrated during the COVID-19 pandemic and other outbreaks (e.g., the 2022 mpox outbreak and 2023 Oropouche fever outbreak; Charniga et al., 2024; Guagliardo et al., 2024). Nowcasting and forecasting provide insightful data for public health decision-makers (Arslan et al., 2020; Charniga et al., 2024). They can evaluate key epidemiological parameters (e.g., case and death counts, hospitalizations, and proper control measures), which are important for preventing virus spread and measuring the risk of reemergence (Birrell et al., 2021; Hinch et al., 2024). In addition, these models allow for the prediction of future health care requirements occasionally up to 2 weeks in advance (Arslan et al., 2020; Steiner et al., 2020).

ML models have become essential for improving the precision of nowcasting and forecasting of viral epidemics and pandemics (Fig. 1 highlights outbreak prediction as a key AI application) (Saleem et al., 2022; Padhi et al., 2023). ML models such as random forest, gradient boosting machines, and GNNs provide robust, data-driven frameworks (Saleem et al., 2022; Hossain et al., 2025; Tripathy et al., 2025). These models exhibited outstanding performance and computational efficiency for nowcasting and allowed for near real-time reporting of epidemiological data (Saleem et al., 2022). Under stable epidemiological conditions with consistent data streams, ML-based forecasting can efficiently incorporate new data to provide early warnings and accurate short-term predictions of case progression (Jiao et al., 2025; Sawhney et al., 2025). However, during abrupt epidemiological shifts—such as the emergence of a novel viral variant, sudden changes in nonpharmaceutical interventions, or superspreading events—predictive performance often declines sharply because the model’s training distribution no longer matches the new reality (Chharia et al., 2024). In addition, ML models can be linked with other models, such as mechanistic or Bayesian time series, to generate robust hybrid systems. These systems can capture the underlying epidemiological processes and the predictive power of data-driven algorithms (Baccega et al., 2024; Hossain et al., 2025; WHO, 2025a). Task-specific models with robust interpretability capabilities and handling of varying forecast lengths were developed from ML models. The flexibility of these models allows them to adapt to local conditions and provide decision-makers with real-time data for better implementation of effective interventions (Fig. 2) (Lee et al., 2018; Chen and Yu, 2025; Liu et al., 2025; Ullanat et al., 2026).

Progression in ML-driven epidemic intelligence involves the analysis of nontraditional, prediagnostic data that detect population behavior and pathogen circulation. In this way, advanced ML models can generate real-time epidemiological data prior to clinical presentation (Sahai et al., 2022). Currently, many signals can be used for nowcasting. These signals include internet search inquiries, social media posts, over-the-counter drug sales, school and work absences, veterinary clinic visits, and viral RNA concentrations in wastewater (Shu and McCauley, 2017; Meijers et al., 2025; Wang et al., 2025). ML models such as multimodal transformers and GNNs have become fundamental for recognizing weak and noisy signals from a variety of heterogeneous sources. They utilize these signals and translate them into calibrated nowcasts and short-term forecasts (Fig. 2; Fritz et al., 2022; Ma et al., 2025). In addition, current DL models can predict clinical case incidence 7–18 days in advance with high associations in well-sampled cities (Chae et al., 2018; Li et al., 2022; Grover et al., 2025; Yang et al., 2026).

The data streams or signals used for nowcasting are not perfect and have limitations. For example, rural regions lack wastewater data, internet search inquiries can be skewed by media spikes, and privacy concerns may restrict mobility data (Akintola et al., 2025). To overcome these issues, multimodal fusion models jointly process diverse sources, including wastewater RNA, Google/Apple mobility reports, X (Twitter) geotagged posts, internet search volume, and electronic health record (EHR) signals (Pang et al., 2023; Babanejaddehaki et al., 2025; Kaur and Butt, 2025). Through the combination of these multiple inputs, fusion models consistently achieve a 25–40% improvement in nowcasting accuracy in comparison with any individual source alone (Roche et al., 2011; Xu et al., 2025). Furthermore, advanced platforms such as EPIWATCH, HealthMap/AI, and the WHO EIOS initiative have the ability to scan over 100,000 online sources daily (Williams and Zhan, 2023; WHO, 2025b). These systems use large language models programmed and trained for outbreak detection and shorten the detection-to-response time from months to a few days (Mangili et al., 2015; Pun et al., 2023). This outstanding rapid processing of data effectively turns epidemic intelligence from a retrospective exercise into a genuinely predictive capability.

Despite these advances, several fundamental challenges remain underappreciated. First, most nowcasting models capture correlations, not causality. A spike in social media posts or wastewater RNA may correlate with subsequent cases but does not prove a causal chain, and interventions based on noncausal signals can be ineffective (Coenen and Weitz, 2018; Mykytyn et al., 2023; Chochlakis et al., 2025). Second, uncertainty quantification is often inadequate; many models report point forecasts without prediction intervals, and the inherent noise in digital data streams is rarely propagated through to final estimates (Holcomb et al., 2024; Kong et al., 2024). Third, algorithmic bias is pervasive: models trained on data from high-income, urban settings fail to generalize to rural or low-resource regions where surveillance infrastructure differs (Orihuel et al., 2021; Yang et al., 2023; Joseph, 2025). Fourth, integration with virological biology remains weak—early warning signals are rarely linked to specific viral lineages, mutation fitness, or immune escape mechanisms, limiting their utility for targeted public health responses (Figgins and Bedford, 2025; Koemen et al., 2026). Addressing these challenges requires closer collaboration between data scientists, virologists, and public health practitioners, as well as the development of causally informed models, rigorous uncertainty reporting, and routine bias audits. This integration of multimodal, real-time data streams into nowcasting and early warning systems represents the second pillar of PV, enabling proactive outbreak detection weeks ahead of clinical surges.

Transmission Dynamics and Forecast Modeling: Beyond Classical Compartmental Models

The third pillar of PV advances transmission dynamics and forecast modeling beyond classical compartmental models through hybrid AI-mechanistic approaches, including neural differential equations. Before the advent of AI models, two conventional compartmental models, susceptible–infectious–recovered and susceptible–exposed–infectious–recovered (SEIR), were commonly used to estimate key epidemiological parameters and to predict viral epidemic trajectories (Hethcote, 2000; Kerr et al., 2021). These conventional models did not perform well during the recent COVID-19 pandemic or with highly transmissible influenza strains. This is attributed to the inherent limitations of these models, which include (i) the assumption of homogeneous mixing and (ii) the inability to detect fine-scale spatial, temporal, and social heterogeneity (Britton et al., 2020; Mollentze and Streicker, 2023; Sun et al., 2025). Subsequently, agent-based models (ABMs) were developed to overcome these limitations. ABMs simulate the behavior and interactions of individual agents (people) within a distinct environment to explicitly model complex contact networks, individual-level risk factors, and the local impact of nonpharmaceutical interventions (NPIs; Ketu and Mishra, 2021; O’Brien et al., 2021; Chen et al., 2023; Kirschbaum et al., 2025). ABMs have been successfully implemented to investigate the epidemiological and evolutionary dynamics of influenza viruses (Rodríguez et al., 2024). Multiscale ABMs have also been used to evaluate the impact of vaccination and NPIs during the COVID-19 pandemic (Kumar and Veer, 2021).

The introduction of AI models (e.g., ML and DL) has revolutionized the field of data-driven and nonparametric forecasting and overcome the limitations of conventional models that require extensive and uncertain parameterization (Satu et al., 2021; Alali et al., 2022; Nitzsche and Simm, 2024). The remarkable increase in epidemiological data streams, including case counts, hospitalization rates, and mobility data, has driven the adoption of advanced statistical and ML models for short-term viral forecasting (Chen and Yu, 2025). Nonparametric methods such as methods of analysis have attracted attention because of their ability to process historical time series for epidemic forecasting (Alali et al., 2022; Sahai et al., 2022; Kim et al., 2025). Method of analogues was originally applied to the spread of influenza epidemics (Villanueva-Flores et al., 2025) and was recently adapted for the prediction of SARS-CoV-2 outbreaks. In addition, DL architectures (e.g., long short-term memory networks) are now frequently utilized to analyze complex and nonlinear time series data. In this way, they offer high-accuracy nowcasting of viral spread by identifying subtle patterns that mechanistic models often overlook (Alassafi et al., 2022). These data-centric approaches are of particular interest during the early phases of a novel viral outbreak when biological parameters are not yet known.

The most promising approach is the development of hybrid models that combine the explanatory power of compartmental models with the predictive accuracy of ML (Ala’raj et al., 2021; Delli Compagni et al., 2022; Jhutty and Hernandez-Vargas, 2022; Cheng et al., 2025; Yousaf et al., 2025). Such models work by utilizing the mechanistic component (e.g., an SEIR model) to capture the fundamental biological processes of viral infection and transmission while employing ML or Bayesian approaches to dynamically estimate time-varying parameters (Delli Compagni et al., 2022; Watson et al., 2023). For example, DL models have been used to forecast time-varying transmission parameters, which are then incorporated into differential equation models to predict future case counts (Kou et al., 2021; Cheng et al., 2025). Similarly, Bayesian hierarchical models integrate diverse data sources and allow for the estimation of the true number of infections by linking observed data (e.g., deaths and hospitalizations) to the underlying transmission process (Reich et al., 2019). This interactive approach represents the current state of the art, providing both a theoretical understanding of viral dynamics and a highly accurate tool for accurate forecasting. These hybrid AI-mechanistic transmission models represent the current state of the art and form the third pillar of PV. By integrating the explanatory power of mechanistic compartmental models with the predictive strength and adaptability of ML, they effectively overcome many limitations of classical compartmental models while capturing spatial, temporal, and social heterogeneity in real time.

Antiviral Discovery and Vaccine Design Accelerated by DL

The fourth pillar of PV harnesses DL to accelerate antiviral discovery, vaccine design, and broad-spectrum countermeasure development. The primary step in the development of viral vaccines and antiviral drugs is to recognize the structures of the target viral protein(s). Determining the immunogenic conformations of viral proteins allows for antigen stabilization and ensures the induction of robust immune responses. In addition, structural recognition facilitates the design of effective antiviral drugs by revealing binding pockets and mechanisms to inhibit key viral functions (Kupperman et al., 2022; Handa et al., 2025; Kumar et al., 2025). Traditional methods such as crystallography or cryo-EM take months to years to determine a high-resolution viral protein structure. Furthermore, these methods can struggle with highly flexible glycoproteins such as influenza HA and the SARS-CoV-2 spike protein in the prefusion conformation, often requiring stabilizing mutations. DL and ML techniques essentially transform the landscape of antiviral drug discovery (Fig. 1 illustrates AI’s role in drug and vaccine development), offering advantages over the time-consuming and costly nature of traditional methods (Varadi et al., 2022). Recently, models such as AlphaFold2 and RoseTTAFold (for monomer and complex prediction), followed by AlphaFold3 (which additionally predicts interactions with nucleic acids, small molecules, and ions), have revolutionized the field of protein structure prediction. These models achieve near-experimental accuracy on CASP14 targets via DL with multiple sequence alignments and end-to-end training (Jumper et al., 2021).

RoseTTAFold introduced a three-track network integrating 1D sequence, 2D distance/orientation, and 3D coordinate representations, allowing rapid and accurate predictions of monomer structures and protein–protein complexes (Table 2; Baek et al., 2021). AlphaFold 3 further extended this ability to the joint prediction of biomolecular complexes, including proteins, nucleic acids, ligands, and ions, outperforming specialized docking and interaction tools (Abramson et al., 2024). SPIRED is another interesting and faster model that introduces a lightweight single-sequence model. It has ∼5-fold faster inference, significantly reduces training costs, and achieves comparable accuracy to OmegaFold and ESMFold on the CAMEO and CASP15 benchmarks (Chen et al., 2024). Furthermore, SPIRED-Fitness was developed by integrating SPIRED into an end-to-end framework. This model facilitates rapid, simultaneous prediction of protein structure and mutational fitness effects, with its derivative SPIRED-Stab reaching state-of-the-art performance in predicting mutational effects on protein stability (Chen et al., 2024).

Table 2.

Comprehensive Summary of Machine Learning and Deep Learning Applications in Antiviral Discovery and Vaccine Design

Application	Deep learning model/Method	Virus/Target	Finding/Outcome	Reference
Antiviral drugs
Drug repurposing	Kernel ridge regression (KRR)	SARS-CoV-2 main protease (Mpro)	Predicted and experimentally validated bromocriptine Half Maximal Inhibitory Concentration (IC₅₀) = 0.13 μM as a potent Mpro inhibitor	(Iype et al., 2022)
De novo drug design	Generative models Chemistry42 platform	SARS-CoV-2 Mpro	De novo design of ISM3312 IC₅₀ = 14 nM, a broad-spectrum covalent Mpro inhibitor that successfully completed Phase I clinical trials	(Suri and Dakshanamurthy, 2022)
Target identification	Deep message passing neural network (D-MPNN)	SARS-CoV-2 Cathepsin L (CTSL)	Identified leupeptin hemisulfate IC₅₀ = 5.77 nM as a highly potent inhibitor of the host protease CTSL	(Yang et al., 2021)
Broad-spectrum antiviral	Markov decision process (MDP)/reinforcement learning	Influenza A and B neuraminidase	De novo design of DS-22-inf-021 IC₅₀ = 9.1 μM for influenza B, a novel neuraminidase inhibitor effective in animal models	(Izmailyan et al., 2024)
Drug repurposing	Molecule transformer–drug target interaction (MT-DTI DL) model	SARS-CoV-2 specifically the 3C-like proteinase and replication complex components	Identified atazanavir as the best candidate with 94.94 nM inhibitory potency. Suggested several antiviral drugs, including Kaletra lopinavir/ritonavir, for SARS-CoV-2 treatment	(Beck et al., 2020)
Drug repurposing and binding dynamics	AI-driven screening DeepRepurpose + molecular dynamics	Monkeypox virus (MPXV)	Identified elvitegravir as a potential inhibitor of MPXV with a strong binding energy of −5.592 kcal/mol	(Pinto et al., 2025)
Drug repurposing/Antiviral prediction	CNN model part of the MPXV-Pred framework	Monkeypox virus (MPXV)	Proposed five approved drugs tilorone, valacyclovir, ribavirin, favipiravir, baloxavir marboxil for MPXV treatment, validated by docking studies	(Hashemi et al., 2024)
Antigen selection, epitope prediction, adjuvant identification, vaccine optimization	ML, DL, HMM, virtual screening, molecular docking, SAR models, generative models, molecular dynamics	Multiple pathogens	Streamlined antigen identification; high-accuracy epitope prediction; novel adjuvant discovery; immunogen stability enhancement; reduced trial duration	(Omar et al., 2025)
Antiviral drugs
Multimodal antiviral discovery	ML model trained on sparse data; unsupervised learning	Human enterovirus 71 (EV71)	Developed an ML framework for virucidal compound prediction. Five out of 8 predicted compounds from the Chinese cosmetic list showed virucidal activity against EV71	(Cesaro et al., 2025)
Antiviral peptide discovery	Graph neural network (GNN), random forest (RF)	Porcine reproductive and respiratory syndrome virus (PRRSV)	RF model achieved the highest Area Under the Curve (AUC) 0.95 ± 2 for predicting antiviral peptides. Highlighted the importance of lysine, arginine, and leucine in prediction	(Zai et al., 2025)
Antiviral drug discovery	Deep graph neural network (GNN) with residual graph convolutional layers	Coronaviruses SARS-CoV-2, MERS-CoV, SARS-CoV	Identified PKI-179 and MTI-31 as potent, broad-spectrum antivirals against coronaviruses, outperforming rapamycin	(Varadi et al., 2022)
Antiviral prediction/drug repurposing	DRaW DL model—CNN architecture	COVID-19 SARS-CoV-2	DRaW outperforms matrix factorization methods in predicting COVID-19 antivirals, with docking studies confirming potential efficacy	(Hashemi et al., 2023)
Antiviral peptide AVP prediction	AI4AVP DL CNN with generative adversarial network (GAN) data augmentation	Antiviral peptides (AVPs)—sequence-based binary classifier	Developed AI4AVP, a deep learning model augmented with GAN data, which outperformed other state-of-the-art AVP predictors	(Lin et al., 2023)
Antiviral peptide design and identification	AVP-GPT (transformer-based language model with multimodal architecture)	Respiratory viruses (RSV, INFVA, SARS-CoV, SARS-CoV-2, HPIV)	Generated 10,000 unique peptides and identified candidates with exceptional potency, including four anti-RSV peptides with Half Maximal Effective Concentration (EC₅₀) values around 0.02 μM	(Zhuang et al., 2025)
Antiviral drug discovery/Predictive modeling	ML models (General)	Yellow fever virus (YFV)	Built predictive models to discover a new antiviral compound against yellow fever virus, validated by synthesis	(Gawriljuk et al., 2021)
Antiviral drug discovery/Virtual screening	Random forest (RF), XGBoost (XGB), SVM, NNET, Naïve Bayes (NB)	SARS-CoV-2 (and variants)	Identified azeliragon (IC₅₀ of 7.68 μM) and other compounds as promising anti-SARS-CoV-2 agents, demonstrating the model’s high predictive power	(Yan et al., 2023)
Antiviral drugs
Viral drug target identification	Structure- and DL-based approach (general)	Viral proteins (general, spurred by COVID-19)	Proposed a structure- and deep learning-based approach to efficiently identify vulnerable regions in viral proteins corresponding to drug binding sites	(Puca and Trent, 2020)
Vaccine design
Multiepitope vaccine design	DeepVacPred in silico DL framework	SARS-CoV-2 spike protein	Computational framework predicted 26 potential vaccine subunits and designed a 694aa multiepitope vaccine construct	(Yao et al., 2023)
T-cell epitope prediction	MUNIS DL model	General pathogens HLA-I epitopes	Highly accurate prediction of CD8 + T cell epitopes, a critical step for rational vaccine design	(Wu et al., 2025)
Protein structure prediction	AlphaFold deep neural network	General proteins	Revolutionized structural vaccinology by predicting high-accuracy 3D protein structures for antigen design	(Viboud et al., 2003)
Multivalent epitope vaccine design	IntegralVac ML, integrating DeepVacPred, MHCSeqNet, and HemoPI	COVID-19 epitopes	A comprehensive ML method that predicts single and multivalent cancer and COVID-19 epitopes with high accuracy 90.1% AUC for immunogenicity	(Thadani et al., 2023)
Protein structure prediction, immunogen design, epitope discovery	AlphaFold, deep neural networks	RSV, SARS-CoV-2, HIV-1, MERS-CoV, SARS-CoV	Structure prediction accuracy equivalent to experimental methods; Stabilized immunogenic structures (RSV F, SARS-CoV-2 spike S2P/HexaPro, HIV BG-SOSIP); novel epitopes identified	(Hederman and Ackerman, 2023)
Multiomic epitope mapping, clinical trial simulation	Random forest, SVM, gradient boosting, CNN, RNN	COVID-19, Zika, Ebola	Reduced discovery timeline by months; optimized trial design and immune prediction; cold-chain logistics optimization	(El Arab et al., 2025)
T-cell epitope identification, B-cell epitope identification, vaccine construct development	CNN, deep neural networks, transformers	SARS-CoV-2	Rapid epitope selection; automated epitope mapping; cost-effective construct development	(Bhattacharya et al., 2025)
Vaccine design
Enhanced antigen prediction, rapid vaccine development	Protein language models, geometric DL (PLGDL)	Mpox, multiple pathogens	Identified known antigens + novel G10R antigen; high-performance screening tool; synergistic sequence-structure analysis	(Zhang et al., 2025)
B-cell epitope discovery, T-cell epitope discovery, immunogen design, cross-variant immunity	ML (supervised/unsupervised), DL	SARS-CoV-2, coronaviruses	Accelerated target selection; reduced experimental burden; optimized antigen immunogenicity; broadly protective epitope prediction	(Bravi, 2024)
Vaccine immune response prediction, genotype–phenotype mapping, drug resistance mapping	XGBoost, random forest, neural networks, SVM, AdaBoost, CNN, DNN	Influenza, RSV, HIV, SARS-CoV-2	Area Under the Receiver Operating Characteristic Curve (AUROC) 0.86 (influenza response); viral trait mapping; resistance pattern prediction; clinical decision support	(Loguercio et al., 2023)
Epitope prediction, structure-integrated epitope prediction	CNN, transformers, graph neural networks (MUNIS, GraphBepi, GearBind)	Viral proteomes, Epstein-Barr Virus (EBV)	87.8% accuracy (AUC = 0.945); 26% improvement over prior methods; 17-fold binding affinity increase; broad-spectrum neutralization maintained	(Wang et al., 2023)
Adjuvant selection, immune response prediction	DL, random forest, RNA transcriptomics	Human Papillomavirus (HPV) (9-valent vaccine)	DL > statistical methods; early-stage RNA expression-based antibody response prediction; gene set differentiation for mechanism insights	(Ji et al., 2025)
Viral immunogenicity prediction, protective antigen identification	Random forest, XGBoost, MLP, SVM, Naive Bayes	31 viral species	Sensitivity 0.966, specificity 0.943, MCC 0.888; superior to VaxiJen 2.0; multispecies antigen identification	(Doneva and Dimitrov, 2024)
Protein subunit vaccine design, viral vector optimization, mRNA vaccine optimization	ML, DL, protein language models, generative models, LinearDesign	SARS-CoV-2, general viral vaccines	De novo immunogenic subunit design; pre-existing immunity mitigation; mRNA stability improvement; SKYCovione™ approved (SK 2022, UK 2023)	(Zhao and Song, 2024)

The emergence of the COVID-19 pandemic motivated the broad implementation of ML, particularly in drug repurposing strategies (Wohlwend et al., 2025). DL models have successfully assisted in screening already approved drug candidates against multiple viral targets (Table 2; Ye et al., 2025). These targets include virus cell entry mechanisms, signal transduction (Luo et al., 2025), and the Mpro protease (Patel et al., 2023). For example, a deep GNN trained on extensive data, including SARS-CoV-2 cell-based assays, was developed. This network, integrated with an in vitro validation approach, was used to prioritize compounds for testing. The outcomes of this work allowed the identification of two compounds, PKI-179 and MTI-31, with broad-spectrum antiviral activity against SARS-CoV-2 variants (Varadi et al., 2022). Another interesting ML-based approach is the deep reinforcement learning-based molecule design platform. This model was successfully used to generate a diverse set of compounds targeting the NA of influenza viruses, which has been shown to be efficacious against Influenza A Virus (IAV) and Influenza B Virus (IBV) both in vitro and in vivo (Izmailyan et al., 2024). This collective evidence emphasizes the potential of AI-driven approaches to speed up drug discovery and enhance global pandemic preparedness (Varadi et al., 2022).

The application of AI/ML extends to the development of novel therapeutics, such as antiviral peptides (AVPs), and to the streamlining of vaccine development and clinical assessment (Table 2). The discovery of AVPs has been significantly advanced by DL, moving beyond conventional high-throughput screening (Duy and Srisongkram, 2025). A hybrid framework combining a Wasserstein generative adversarial network with gradient penalty and a bidirectional long short-term memory (BiLSTM) network was specifically developed to design and classify newly generated AVPs, successfully identifying 815 novel candidates across 12 distinct viral endpoints (Duy and Srisongkram, 2025). This generative approach complements the use of AI/ML in vaccine development, such as the technologies utilized for the AstraZeneca/Oxford University vaccine (AZD1222/Covishield; AstraZeneca, 2025; Kumar et al., 2025). In the clinical context, AI/ML has been employed during clinical trials for event assessment. This resulted in an improvement in the process at various stages and thereby decreased the total duration of the trials (AstraZeneca, 2025; Kumar et al., 2025). Furthermore, ML has been used to analyze EHRs, which are instrumental in optimizing clinical trial patient recognition and recruitment (Cai et al., 2021; Olawade et al., 2024). In addition to clinical logistics, graphical-based knowledge and image analysis have also been employed to gain new insights into illnesses. In some studies, these AI tools have demonstrated up to 30% faster biomarker detection compared with human pathologists, depending on the specific metric and evaluation setting (Haddad et al., 2021; Li et al., 2024; Shah et al., 2024; Tulchinsky et al., 2025).

Despite the demonstrated successes in both drug and vaccine pipelines, the implementation of ML in antiviral research faces several challenges that require careful consideration (Anokian et al., 2024; MacIntyre et al., 2023). One major limitation of drug repurposing is that various antiviral candidates lack proper validation in physiologically relevant models. This aligns with the fact that several repurposed drugs failed to demonstrate clinical benefits during the COVID-19 pandemic. Therefore, robust in vitro and preclinical validation before considering the off-label use of drug candidates is a priority. Another limitation is the low amount of data available per target for quantitative structure–activity relationship analysis. This can lead neural networks to learn to remember training examples rather than discovering a general predictive algorithm (Varadi et al., 2022). Although the BiLSTM model performs well in AVP prediction, more advanced models, such as PreTP-Stack (Yang et al., 2024) and DeepAVP (Li et al., 2025), have superior performance. The superior predictive performance of these models highlights the need for further optimization of the BiLSTM approach (Duy and Srisongkram, 2025). Therefore, AI-driven approaches provide significant advantages in terms of rapid efficiency and broad applicability across diverse viral targets. However, rigorous caution and experimental validation are essential to confirm the clinical relevance and safety of the predicted results. The generative, structure prediction, and ML approaches described here constitute the fourth pillar of PV. By accelerating antiviral discovery and vaccine design from years to months, they provide a powerful and distinctive capability within epidemic intelligence frameworks. Rigorous experimental validation remains essential to translate these computational advances into safe and effective countermeasures. The fifth pillar, ecological zoonotic risk assessment, is discussed within the context of the “The Emerging Discipline of PV: Definitions, Scope, and Imperatives” section and Figure 2.

Ethical, Governance, and Equity Challenges in PV

The application of AI-based models in PV faces profound ethical, governance, and equity challenges. Predictive models often rely on extensive datasets integrating genomic, epidemiological, social media sources, and so on. This raises significant privacy concerns and risks of surveillance overreach, especially when applied to contact tracing or population monitoring during viral outbreaks (Borda et al., 2022; Meng et al., 2024). Since AI models depend mainly on the input data for prediction, algorithmic biases can emerge from incomplete or skewed training data. This can lead to inaccurate predictions and perpetuate global health disparities (Cesaro et al., 2025; Pagsuyoin et al., 2025). Regulatory systems are lagging rapidly behind AI progress, as issues of transparency in computationally complex models hinder accountability and explainability. In addition, they struggle to address the difficulties of international cooperation and accountability in managing cross-border viral threats (Jit et al., 2021; de Lima and Quaresma, 2025; Ding et al., 2025). Unequal access to AI technologies and data infrastructure intensifies equity challenges. This inequality widens the gap in outbreak preparedness between high-resource and resource-limited settings. Consequently, there is a risk of “digital colonization,” where data from affected or vulnerable regions are used to feed models that primarily benefit wealthier nations (Borda et al., 2022). Addressing these challenges requires the development of integrative ethics frameworks that involve multidisciplinary stakeholders. Furthermore, robust data governance, with a strong emphasis on fairness and inclusiveness, is essential. Last, these efforts must be supported by policies that promote open-source models and privacy-preserving techniques, ensuring that PV can advance public health without amplifying inequalities (Shahid et al., 2020; van der Horst et al., 2024; Zhao and Zai, 2025).

Conclusions

The convergence of high-throughput genomic surveillance, open data ecosystems, and exponential advances in AI has given rise to PV as a transformative discipline, shifting traditional reactive virology toward proactive forecasting and preemptive intervention. Advanced tools such as pLMs, multimodal early-warning systems, neural differential equations for transmission dynamics, and generative molecular design have already delivered notable improvements in predicting mutational fitness, antigenic evolution, cryptic transmission clusters, and zoonotic spillover risk—often outperforming decades of conventional laboratory and epidemiological approaches when benchmarked against real-world SARS-CoV-2 variants, seasonal influenza, and emerging outbreaks. These capabilities demonstrate that meaningful foresight months to years ahead is now technically feasible. However, the full realization of PV’s operational impact hinges on addressing several critical challenges. The central challenge for PV is no longer whether we can predict viral threats but how we can predict responsibly and equitably. Key questions that will define the next decade of the field include enhancing mechanistic interpretability of opaque “black-box” models, seamlessly integrating heterogeneous real-time data streams from diverse global settings (particularly from low- and middle-income countries [LMICs]), rigorously validating in silico predictions in physiologically relevant biological systems, and developing robust governance frameworks to prevent surveillance overreach and mitigate the risk of exacerbating global health disparities. By confronting these challenges head-on through explainable AI (XAI), federated data infrastructure, pan-viral countermeasure design, and equitable international collaboration, PV can fulfill its promise of strengthening global health security and transforming pandemic preparedness from reactive crisis management into proactive defense.

Future Directions

To fully realize the potential of PV, future efforts must prioritize four critical directions. First, the field must move beyond opaque “black-box” models by advancing XAI techniques that provide mechanistic interpretability, thereby enhancing trust, robustness, and generalizability—particularly for truly novel, out-of-distribution viruses. Promising early applications, such as SHapley Additive exPlanations (SHAP)-integrated models for prognostic severity prediction in COVID-19 and interpretable DL frameworks that reveal key mutation contributions in HIV drug resistance prediction, demonstrate how XAI can bridge the gap between predictive accuracy and biological insight; expanding these approaches across mutational fitness, antigenic escape, and zoonotic risk models will be essential. Second, a federated, globally standardized data infrastructure is required to enable seamless, real-time fusion of heterogeneous genomic, clinical, environmental, and mobility data streams, with deliberate expansion of high-quality surveillance efforts in LMICs to reduce predictive bias and improve representativeness. Third, successes in in silico design should be leveraged to accelerate generative AI for pan-viral countermeasures, with a focus on developing broad-spectrum antivirals and vaccines capable of preemptively neutralizing entire viral clades and thereby compressing the timeline from prediction to clinical readiness. Last, operational deployment and equitable governance represent the most pressing overarching challenge. Beyond developing user-friendly alert systems that integrate PV outputs into public health workflows, the field urgently needs new international treaties or targeted amendments to the International Health Regulations that explicitly address the unique issues raised by AI-generated predictions—including accountability for model-driven decisions, mechanisms for cross-border validation of forecasts, standardized protocols for data sharing and bias mitigation, and enforceable safeguards against surveillance overreach and “digital colonization.” Establishing such governance architectures through multidisciplinary collaboration among scientists, ethicists, policymakers, and global health bodies will ensure that the profound predictive power of PV benefits all nations equitably and strengthens rather than fragments global health security.

Author’s Contributions

M.A.F. conceived the idea, conducted the literature review, drafted the entire manuscript, prepared the tables and figures, and revised the final version. The author read and approved the final manuscript.

Footnotes

Acknowledgments

The author would like to acknowledge the use of Grok (built by xAI, version 4.1, accessed in April 2026) during the preparation of this article. Grok was used solely for limited purposes of text structuring, language editing, and phrasing suggestions. All content was reviewed, revised, and finalized by the authors, who take full responsibility for the accuracy, originality, and integrity of the published work.

Author Disclosure Statement

The author has no relevant financial or nonfinancial interests to disclose.

Funding Information

The author declares that no funds, grants, or other support were received during the preparation of this article.

References

Aassve

, Alfani

, Gandolfi

, Le Moglie

. Epidemics and trust: The case of the Spanish Flu. Health Econ 2021;30(4):840–857.

Abbott

SHJ

, Thompson

, et al. Stimating the time-varying reproduction number of SARS-CoV-2 using national and subnational case counts [version 2; peer review: 1 approved, 1 approved with reservations]. Wellcome Open Res 2020;5:112.

Abramson

, Adler

, Dunger

, et al. Accurate structure prediction of biomolecular interactions with AlphaFold 3. Nature 2024;630(8016):493–500.

Akintola

, Tchoumo

, Abdulrauf

, et al. Machine learning for early disease outbreak detection in the United States: An integrative systematic review of multisource data integration. Int J Trop Dis Health 2025;46(10):52–67.

Ala’raj

, Majdalawieh

, Nizamuddin

. Modeling and forecasting of COVID-19 using a hybrid dynamic model based on SEIRD with ARIMA corrections. Infect Dis Model 2021;6:98–111.

Alali

, Harrou

, Sun

. A proficient approach to forecast COVID-19 spread via optimized dynamic machine learning models. Sci Rep 2022;12(1):2467.

Alassafi

, Jarrah

, Alotaibi

. Time series predicting of COVID-19 based on deep learning. Neurocomputing (Amst) 2022;468:335–344.

Ali Maher

, Pichierri

, Popescu

, Bellizzi

. Artificial intelligence and public health data in Emergencies: A critical issue in the middle east. Am J Trop Med Hyg 2026;114(4):606–608.

Anokian

, Bernett

, Freeman

, et al. Machine learning and artificial intelligence in drug repurposing—challenges and perspectives. DrugRepo 2024;1.

10.

Arslan

, Ozdemir

, Ucar

. Nowcasting and forecasting the spread of COVID-19 and healthcare demand in Turkey, a modeling study. Front Public Health 2020;8:575145.

11.

AstraZeneca. (2025) Data Science & Artificial Intelligence: Unlocking new science insights. AstraZeneca.

12.

Babanejaddehaki

, An

, Papagelis

. Disease outbreak detection and forecasting: A review of methods and data sources. ACM Trans Comput Healthcare 2025;6(2):1–40.

13.

Baccega

, Castagno

, Fernández Anta

, Sereno

. Enhancing COVID-19 forecasting precision through the integration of compartmental models, machine learning and variants. Sci Rep 2024;14(1):19220.

14.

Baek

, DiMaio

, Anishchenko

, et al. Accurate prediction of protein structures and interactions using a three-track neural network. Science 2021;373(6557):871–876.

15.

Baig

, Hobson

, GholamHosseini

, et al. Generative AI in improving personalized patient care plans: Opportunities and barriers towards its wider adoption. Applied Sciences 2024;14(23):10899.

16.

Bakas

, Li

, Shah

, Roth

. Federated learning in healthcare: From research to real-world deployment. Annu Rev Biomed Eng 2026;28(1):163–186.

17.

Bartlow

, Manore

, Xu

, et al. Forecasting zoonotic infectious disease response to climate change: Mosquito vectors and a changing environment. Vet Sci 2019;6(2):40.

18.

Beck

, Shin

, Choi

, et al. Predicting commercially available antiviral drugs that may act on the novel coronavirus (SARS-CoV-2) through a drug-target interaction deep learning model. Comput Struct Biotechnol J 2020;18:784–790.

19.

Bedford

, Greninger

, Roychoudhury

, Seattle Flu Study Investigators. et al. Cryptic transmission of SARS-CoV-2 in Washington state. Science 2020;370(6516):571–575.

20.

Bhattacharya

, Lo

Y-H

, Chatterjee

, et al. Deep learning in next-generation vaccine development for infectious diseases. Mol Ther Nucleic Acids 2025;36(3):102586.

21.

Birrell

, Blake

, van Leeuwen

, et al. Real-time nowcasting and forecasting of COVID-19 dynamics in England: The first wave. Philos Trans R Soc Lond B Biol Sci 2021;376(1829):20200279.

22.

Blanco-Melo

, Campbell

, Zhu

, et al. A novel approach to exploring the dark genome and its application to mapping of the vertebrate virus fossil record. Genome Biol 2024;25(1):120.

23.

Blassel

, Tostevin

, Villabona-Arenas

, UK HIV Drug Resistance Database. et al. Using machine learning and big data to explore the drug resistance landscape in HIV. PLoS Comput Biol 2021;17(8):e1008873.

24.

Borda

, Molnar

, Neesham

, Kostkova

. Ethical issues in AI-Enabled disease surveillance: Perspectives from global health. Applied Sciences 2022;12(8):3890.

25.

Bowyer

, Allen

, Furnham

. Unveiling the ghost: Machine learning’s impact on the landscape of virology. J Gen Virol 2025;106(1):2067.

26.

Brandes

, Goldman

, Wang

, et al. Genome-wide prediction of disease variant effects with a deep protein language model. Nat Genet 2023;55(9):1512–1522.

27.

Bravi

. Development and use of machine learning algorithms in vaccine target selection. NPJ Vaccines 2024;9(1):15.

28.

Britton

, Ball

, Trapman

. A mathematical model reveals the influence of population heterogeneity on herd immunity to SARS-CoV-2. Science 2020;369(6505):846–849.

29.

Cai

, Cai

, Dahal

, et al. Improving the efficiency of clinical trial recruitment using an ensemble machine learning to assist with eligibility screening. ACR Open Rheumatol 2021;3(9):593–600.

30.

Carlson

, Albery

, Merow

, et al. Climate change increases cross-species viral transmission risk. Nature 2022;607(7919):555–562.

31.

Carlson

, Farrell

, Grange

, et al. The future of zoonotic risk prediction. Philosophical Transactions of the Royal Society B: Biological Sciences 2021;376(1837).

32.

Carlson

, Thompson

, Kacmar

. Double crossed: The spillover and crossover effects of work demands on work outcomes through the family. J Appl Psychol 2019;104(2):214–228.

33.

(CDC) CfDCaP. (2024) Ebola outbreaks. CDC.

34.

Cesaro

, Wan

, Shi

, et al. Antiviral discovery using sparse datasets by integrating experiments, molecular simulations, and machine learning. Cell Rep Phys Sci 2025;6(5):102554.

35.

Chadaga

, Prabhu

, Sampathila

, et al. Explainable artificial intelligence approaches for COVID-19 prognosis prediction using clinical markers. Sci Rep 2024;14(1):1783.

36.

Chae

, Kwon

, Lee

. Predicting infectious disease using deep learning and big data. Int J Environ Res Public Health 2018;15(8):1596.

37.

Charniga

, Madewell

, Masters

, et al. Nowcasting and forecasting the 2022 U.S. mpox outbreak: Support for public health decision making and lessons learned. Epidemics 2024;47:100755.

38.

Chen

, Yu

. A transferable machine learning model for real-time forecast of epidemic dynamics and pre-trigger event warning. AI Civ Eng 2025;4(1):18.

39.

Chen

, Fan

, Lu

, et al. Intelligence on the graph: Graph neural networks for mechanistic drug target discovery. J Pharm Anal 2025:101508.

40.

Chen

, Jiang

, Li

, Zhou

. A stochastic agent-based model to evaluate COVID-19 transmission influenced by human mobility. Nonlinear Dyn 2023;111:1–17.

41.

Chen

, Xu

, Liu

, et al. An end-to-end framework for the prediction of protein structure and fitness from single sequence. Nat Commun 2024;15(1):7400.

42.

Cheng

, Aruchunan

, Noor Aziz

. Leveraging dynamics informed neural networks for predictive modeling of COVID-19 spread: A hybrid SEIRV-DNNs approach. Sci Rep 2025;15(1):2043.

43.

Cheng

, Mao

, Tang

, et al. Zero-shot prediction of mutation effects with multimodal deep representation learning guides protein engineering. Cell Res 2024;34(9):630–647.

44.

Chharia

, Jeevan

, Jha

, et al. Accuracy of US CDC COVID-19 forecasting models. Front Public Health 2024;12:1359368.

45.

Chochlakis

, Tzedakis

, Kokkinomagoula

, et al. Challenges on the implementation of wastewater-based epidemiology as a prediction tool: The paradigm of SARS-CoV-2. Sci Total Environ 2025;981:179593.

46.

Coenen

, Weitz

. Limitations of correlation-based inference in complex virus-microbe communities. mSystems 2018;3(4):e00084–e00018.

47.

Cutler

, Summers

. The COVID-19 pandemic and the $16 Trillion virus. JAMA 2020;324(15):1495–1496.

48.

de Lima

, Quaresma

JAS

. Emerging technologies transforming the future of global biosecurity. Front Digit Health 2025;7:1622123.

49.

Delli Compagni

, Cheng

, Russo

, Van Boeckel

. A hybrid Neural Network-SEIR model for forecasting intensive care occupancy in Switzerland during COVID-19 epidemics. PLoS One 2022;17(3):e0263789.

50.

Ding

, Shang

, Xie

, et al. Artificial intelligence in the COVID-19 pandemic: Balancing benefits and ethical challenges in China’s response. Humanit Soc Sci Commun 2025;12(1):245.

51.

Doneva

, Dimitrov

. Viral immunogenicity prediction by machine learning methods. Int J Mol Sci 2024;25(5):2949.

52.

, King

, Woods

, Pascual

. Evolution-informed forecasting of seasonal influenza A (H3N2). Sci Transl Med 2017;9(413):eaan5325.

53.

Duy

, Srisongkram

. A hybrid framework of generative deep learning for antiviral peptide discovery. Sci Rep 2025;15(1):25554.

54.

Ebrahimi

, Ghaemi

. Artificial intelligence and spillover prediction: Aligning innovation with empirical reality in One Health surveillance. Lancet Infect Dis 2026;26(1):e6.

55.

El Arab

, Alkhunaizi

, Alhashem

, et al. Artificial intelligence in vaccine research and development: An umbrella review. Front Immunol 2025;16:1567116.

56.

FAO FaAOotUN. (2025) Breaking down silos: Why the One Health approach is our best defence against future pandemics. FAO.

57.

Feng

, Chen

, Wang

, et al. Trends and impacts of SARS-CoV-2 genome sharing: A comparative analysis of China and the global community, 2020–2023. Front Public Health 2024;12:1491623.

58.

Figgins

, Bedford

. Frequency dynamics predict viral fitness, antigenic relationships and epidemic growth. medRxiv 2025.

59.

Findlater

, Bogoch

. Human Mobility and the Global Spread of Infectious Diseases: A Focus on Air Travel. Trends Parasitol 2018;34(9):772–783.

60.

Flores

, Kim

, Young

. Addressing bias in artificial intelligence for public health surveillance. J Med Ethics 2024;50(3):190–194.

61.

Fritz

, Dorigatti

, Rügamer

. Combining graph neural networks and spatio-temporal disease models to improve the prediction of weekly COVID-19 cases in Germany. Sci Rep 2022;12(1):3930.

62.

Gao

. Infodemiology: The science studying infodemic and inforus. China CDC Wkly 2022;4(52):1181–1182.

63.

Gawriljuk

, Foil

, Puhl

, et al. Development of machine learning models and the discovery of a new antiviral compound against yellow fever virus. J Chem Inf Model 2021;61(8):3804–3813.

64.

Gibson

, Li

, et al. PRIME: An evaluation framework for protein representation inference and generalization in viral mutation space. BMC Genomics 2026.

65.

Gomes

RAL

, Zerbini

. ConCreT, a 2D convolutional neural network for taxonomic classification applied to viruses in the phylum Cressdnaviricota. J Virol Methods 2023;320:114789.

66.

Gracia Carmona

, Leipart

, Amdam

, et al. Leveraging protein language models and a scoring function for indel characterization and transfer learning. Patterns 2026;7(2):101425.

67.

Grover

, Hill

, Kasarskis

, et al. Identifying real time surveillance indicators to estimate COVID-19 hospital admissions in Colorado during and after the public health emergency. Sci Rep 2025;15(1):21614.

68.

Grange

, Goldstein

, Johnson

, et al. Ranking the risk of animal-to-human spillover for newly discovered viruses. Proc Natl Acad Sci USA [2021;118(15).

69.

Guagliardo

SAJ

, Connelly

, Lyons

, et al. Reemergence of oropouche virus in the Americas and risk for spread in the United States and Its Territories, 2024. Emerg Infect Dis 2024;30(11):2241–2249.

70.

Haddad

, Helgeson

, Pomerleau

, et al. Accuracy of an artificial intelligence system for cancer clinical trial eligibility screening: Retrospective pilot study. JMIR Med Inform 2021;9(3):e27767.

71.

Hadfield

, Megill

, Bell

, et al. Nextstrain: Real-time tracking of pathogen evolution. Bioinformatics 2018;34(23):4121–4123.

72.

Han

, Chen

, Xu

, et al. Predicting the antigenic evolution of SARS-COV-2 with deep learning. Nat Commun 2023;14(1):3478.

73.

Handa

, Saha

, Narayanan

, et al. Structural virology: The key determinants in development of antiviral therapeutics. Viruses 2025;17(3):417.

74.

Hardy-Werbin

, Maiques

, Busto

, et al. MultiCOVID: A multi modal deep learning approach for COVID-19 diagnosis. Sci Rep 2023;13(1):18761.

75.

Hashemi

, Zabihian

, Hajsaeedi

, Hooshmand

. Antivirals for monkeypox virus: Proposing an effective machine/deep learning framework. PLoS One 2024;19(9):e0299342.

76.

Hashemi

, Zabihian

, Hooshmand

, Gharaghani

. DRaW: Prediction of COVID-19 antivirals by deep learning—an objection on using matrix factorization. BMC Bioinformatics 2023;24(1):52.

77.

Hassan Nishan

. Digital epidemiology: Utilizing big data for public health surveillance and disease outbreak prediction. J Public Health Res 2026;15(2):22799036261455523.

78.

, Zhou

, Bai

, et al. ViTax: Adaptive hierarchical viral taxonomy classification with a taxonomy belief tree on a foundation model. Brief Bioinform 2024;26(1):bbaf041.

79.

Hederman

, Ackerman

. Leveraging deep learning to improve vaccine design. Trends Immunol 2023;44(5):333–344.

80.

Hethcote

. The mathematics of infectious diseases. SIAM Rev 2000;42(4):599–653.

81.

Hinch

, Panovska-Griffiths

, Ward

, et al. Quantification of the time-varying epidemic growth rate and of the delays between symptom onset and presenting to healthcare for the mpox epidemic in the UK in 2022. Sci Rep 2024;14(1):19755.

82.

Holcomb

, Christensen

, Hoffman

, et al. Estimating rates of change to interpret quantitative wastewater surveillance of disease trends. Sci Total Environ 2024;951:175687.

83.

Holmes

, Krammer

, Goodrum

. Virology—The next fifty years. Cell 2024;187:5128–5145.

84.

Hopf

, Ingraham

, Poelwijk

, et al. Mutation effects predicted from sequence co-variation. Nat Biotechnol 2017;35(2):128–135.

85.

Hossain

, Goyal

, Martin

, et al. Machine learning and probabilistic approaches for forecasting COVID-19 transmission and cases. medRxiv 2025.

86.

Ito

, Strange

, Liu

, The Genotype to Phenotype Japan C

et al.;. A protein language model for exploring viral fitness landscapes. Nat Commun 2025;16(1):4236.

87.

Islam

, Hasan

, Shajeeb

, et al. Multimodal marvels of deep learning in medical diagnosis using image, speech, and text: A comprehensive review of COVID-19 detection. AI Open 2025;6:12–44.

88.

Iype

, Pillai

, Kumar

, et al. In silico and in vitro assays reveal potential inhibitors against 3CL(pro) main protease of SARS-CoV-2. J Biomol Struct Dyn 2022;40(23):12800–12811.

89.

Izmailyan

, Matevosyan

, Khachatryan

, et al. Discovery of new antiviral agents through artificial intelligence: In vitro and in vivo results. Antiviral Res 2024;222:105818.

90.

Jhutty

, Hernandez-Vargas

. Parameter estimation in hybrid machine learning and mechanistic models of infectious diseases. IFAC Pap OnLine 2022;55(16):178–183.

91.

, Bekkari

, Patel

, et al. Machine learning insights into vaccine adjuvants and immune outcomes. Front Immunol 2025;16:1654060.

92.

Jiao

, Wang

, Ye

, et al. Spatio-temporal epidemic forecasting using mobility data with LSTM networks and attention mechanism. Sci Rep 2025;15(1):9603.

93.

Jit

, Ananthakrishnan

, McKee

, et al. Multi-country collaboration in responding to global infectious disease threats: Lessons for Europe from the COVID-19 pandemic. Lancet Reg Health Eur 2021;9:100221.

94.

Joseph

. Algorithmic bias in public health AI: A silent threat to equity in low-resource settings. Front Public Health 2025;13:1643180.

95.

Jumper

, Evans

, Pritzel

, et al. Highly accurate protein structure prediction with AlphaFold. Nature 2021;596(7873):583–589.

96.

Kaur

, Butt

. AI-driven epidemic intelligence: The future of outbreak detection and response. Front Artif Intell 2025;8:1645467.

97.

Kawasaki

, Suzuki

, Hamada

. Hidden challenges in evaluating spillover risk of zoonotic viruses using machine learning models. Commun Med (Lond) 2025;5(1):187.

98.

Keeling

, Rohani

. (2008) Modeling Infectious Diseases in Humans and Animals. Princeton University Press.

99.

Kerr

, Stuart

, Mistry

, et al. Covasim: An agent-based model of COVID-19 dynamics and interventions. PLoS Comput Biol 2021;17(7):e1009149.

100.

Ketu

, Mishra

. Enhanced Gaussian process regression-based forecasting model for COVID-19 outbreak and significance of IoT for its detection. Appl Intell (Dordr) 2021;51(3):1492–1512.

101.

Kim

, Jung

, Geard

, et al. Realistic mobility integration in agent-based model for targeted public health strategies. Chaos, Solitons & Fractals 2025;199:116725.

102.

Kirschbaum

, Kongkitimanon

, Frank

, et al. VirusWarn: A mutation-based early warning system to prioritize concerning SARS-CoV-2 and influenza virus variants from sequencing data. Comput Struct Biotechnol J 2025;27:1081–1088.

103.

Koemen

, Faria

, Bastos

, Machine Learning & Global Health Network. et al. Fast and trustworthy nowcasting of dengue fever: A case study using attention-based probabilistic neural networks in São Paulo, Brazil. Epidemics 2026;54:100880.

104.

Kong

, Guo

, Lee

C-W

. Enhancing COVID-19 prevalence forecasting: A hybrid approach integrating epidemic differential equations and recurrent neural networks. AppliedMath 2024;4(2):427–441.

105.

Kou

, Wang

, Li

, et al. A multi-scale agent-based model of infectious disease transmission to assess the impact of vaccination and non-pharmaceutical interventions: The COVID-19 case. J Saf Sci Resil 2021;2(4):199–207.

106.

Kumar

, Maji

, Tiwari

, et al. Precision vaccine design targeting the prefusion state of viral glycoproteins: Advances in structural vaccinology. Biochem Pharmacol 2025;242(Pt 2):117349.

107.

Kumar

, Romano

, Ritchie

. CASTER-DTA: Equivariant graph neural networks for predicting drug–target affinity. Brief Bioinform 2025;26(5):bbaf554.

108.

Kumar

, Veer

. How artificial intelligence and internet of things can aid in the distribution of COVID-19 vaccines. Diabetes Metab Syndr 2021;15(3):1049–1050.

109.

Kupperman

, Leitner

, Ke

. A deep learning approach to real-time HIV outbreak detection using genetic data. PLoS Comput Biol 2022;18(10):e1010598.

110.

Lai

, Wulff

, Cao

, et al. An interpretable time series machine learning method for varying forecast and nowcast lengths in wastewater-based epidemiology. MethodsX 2023;11:102382.

111.

Lee

, Huddleston

, Doud

, et al. Deep mutational scanning of hemagglutinin helps predict evolutionary fates of human H3N2 influenza variants. Proc Natl Acad Sci U S A 2018;115(35):E8276–Ee8285.

112.

, Pu

, Tang

, et al. DeepAVP: A dual-channel deep neural network for identifying variable-length antiviral peptides. IEEE J Biomed Health Inform 2020;24(10):3012–3019.

113.

, Yan

, Chu

, et al. A deep learning framework for using search engine data to predict influenza-like illness and distinguish epidemic and nonepidemic seasons: Multifeature time series analysis. J Med Internet Res 2025;27:e71786.

114.

, Huang

, Zitnik

. Graph representation learning in biomedicine and healthcare. Nat Biomed Eng 2022;6(12):1353–1369.

115.

, Li

, Shang

, Kong

. A sequence-based machine learning model for predicting antigenic distance for H3N2 influenza virus. Front Microbiol 2024;15:1345794.

116.

, Wu

, Zhou

. EVEscape: Revealing potential escape sites based on the viral variation landscape. Biophys Rep 2024;10(2):133–134.

117.

Lin

, Karthikeyan

, Satterlund

, et al. Optimizing campus-wide COVID-19 test notifications with interpretable wastewater time-series features using machine learning models. Sci Rep 2023;13(1):20670.

118.

Lin

T-T

, Sun

Y-Y

, Wang

C-T

, et al. AI4AVP: An antiviral peptides predictor in deep learning approach with generative adversarial network data augmentation. Bioinform Adv 2022;2(1):vbac080.

119.

Lin

, Akin

, Rao

, et al. Evolutionary-scale prediction of atomic-level protein structure with a language model. Science 2023;379(6637):1123–1130.

120.

Liu

, Clemente

, Poirier

, et al. Real-Time forecasting of the COVID-19 outbreak in Chinese Provinces: Machine learning approach using novel digital data and estimates from mechanistic models. J Med Internet Res 2020;22(8):e20285.

121.

Liu

, Young

, Lamb

, et al. PLM-interact: Extending protein language models to predict protein-protein interactions. Nat Commun 2025;16(1):9012.

122.

Liu

, Nie

, Si

, et al. Generative prediction of real-world prevalent SARS-CoV-2 mutation with in silico virus evolution. Brief Bioinform 2025;26(3):bbaf276.

123.

Livieratos

, Kagadis

, Gogos

, Akinosoglou

. AI methods tailored to influenza, RSV, HIV, and SARS-CoV-2: A focused review. Pathogens 2025;14(8):748.

124.

Loguercio

, Calverley

, Wang

, et al. Understanding the host-pathogen evolutionary balance through Gaussian process modeling of SARS-CoV-2. Patterns (N Y) 2023;4(8):100800.

125.

Lou

, Liang

, Cao

, et al. Predictive evolutionary modelling for influenza virus by site-based dynamics of mutations. Nat Commun 2024;15(1):2546.

126.

Loucera

, Esteban-Medina

, Rian

, et al. Drug repurposing for COVID-19 using machine learning and mechanistic models of signal transduction circuits related to SARS-CoV-2 infection. Signal Transduct Target Ther 2020;5(1):290.

127.

Luo

, Wang

, Fan

, et al. A novel graph neural network based approach for influenza-like illness nowcasting: Exploring the interplay of temporal, geographical, and functional spatial features. BMC Public Health 2025;25(1):408.

128.

, Wang

, Rathi

, Zhan

. (2025) Chapter Three - Artificial intelligence in the development of antiviral drugs: Progress and applications. In: Feng

. (ed) Annual Reports in Medicinal Chemistry. Academic Press, pp. 87–126.

129.

MacIntyre

, Chen

, Kunasekaran

, et al. Artificial intelligence in public health: The potential of epidemic early warning systems. J Int Med Res 2023;51(3):3000605231159335.

130.

Mangili

, Vindenes

, Gendreau

. Infectious risks of air travel. Microbiol Spectr 2015;3(5); doi: 10.1128/microbiolspec.iol1125-0009-2015

131.

Mastala

, Kalonde

, Chikhola

, et al. Challenges and potential opportunities for improving One Health surveillance in low-resource settings: Insights from rabies surveillance in Malawi. One Health Outlook 2026;8(1):5.

132.

Matson

, Comba

, Silvert

, et al. A deep learning approach predicting the activity of COVID-19 therapeutics and vaccines against emerging variants. NPJ Syst Biol Appl 2024;10(1):138.

133.

McMahan

, Self

, Rennert

, et al. COVID-19 wastewater epidemiology: A model to estimate infected populations. Lancet Planet Health 2021;5(12):e874–e881.

134.

Meijers

, Ruchnewitz

, Eberhardt

, et al. Concepts and methods for predicting viral evolution. Methods Mol Biol 2025;2890:253–290.

135.

Mello

, Wang

. Ethics and governance for digital disease surveillance. Science 2020;368(6494):951–954.

136.

Meng

, Liu

, Song

, et al. PREDAC-CNN: Predicting antigenic clusters of seasonal influenza A viruses with convolutional neural network. Brief Bioinform 2024;25(2):bbae033.

137.

Ming

, Chen

, Wang

, et al. HostNet: Improved sequence representation in deep neural networks for virus-host prediction. BMC Bioinformatics 2023;24(1):455.

138.

Mock

, Viehweger

, Barth

, Marz

. VIDHOP, viral host prediction with deep learning. Bioinformatics 2021;37(3):318–325.

139.

Moein

, Nickaeen

, Roointan

, et al. Inefficiency of SIR models in forecasting COVID-19 epidemic: A case study of Isfahan. Sci Rep 2021;11(1):4725.

140.

Mollentze

, Streicker

. Predicting zoonotic potential of viruses: Where are we? Curr Opin Virol 2023;61:101346.

141.

Morbey

, Todkill

, Watson

, Elliot

. Machine learning forecasts for seasonal epidemic peaks: Lessons learnt from an atypical respiratory syncytial virus season. PLoS One 2023;18(9):e0291932.

142.

Mulomba

, Kiketa

, Kutangila

, et al. Applying causal machine learning to spatiotemporal data analysis: An investigation of opportunities and challenges. IEEE Access 2025;13:141832–141857.

143.

Mykytyn

, Fouchier

RAM

, Haagmans

. Antigenic evolution of SARS coronavirus 2. Curr Opin Virol 2023;62:101349.

144.

Naseer

, Khalid

, Parveen

, et al. COVID-19 outbreak: Impact on global economy. Front Public Health 2022;10:1009393.

145.

Nie

, Liu

, Chen

, et al. A unified evolution-driven deep learning framework for virus variation driver prediction. Nat Mach Intell 2025;7(1):131–144.

146.

Nikparvar

, Rahman

, Hatami

, Thill

J-C

. Spatio-temporal prediction of the COVID-19 pandemic in US counties: Modeling with a deep LSTM neural network. Sci Rep 2021;11(1):21715.

147.

Nitzsche

, Simm

. Agent-based modeling to estimate the impact of lockdown scenarios and events on a pandemic exemplified on SARS-CoV-2. Sci Rep 2024;14(1):13391.

148.

O’Brien

, Raman

, Ellis

, et al. The use of electronic health records for recruitment in clinical trials: A mixed methods analysis of the Harmony Outcomes Electronic Health Record Ancillary Study. Trials 2021;22:465.

149.

Olawade

, Teke

, Fapohunda

, et al. Leveraging artificial intelligence in vaccine development: A narrative review. J Microbiol Methods 2024;224:106998.

150.

Omar

, Soffer

, Agbareia

, et al. Sociodemographic biases in medical decision making by large language models. Nat Med 2025;31(6):1873–1881.

151.

Orihuel

, Sapena

, Navarro-Ortiz

. An empirical algorithm for COVID-19 nowcasting and short-term forecast in Spain: A kinematic approach. ASI 2021;4(1):2; doi: 10.3390/asi4010002

152.

Padhi

, Agarwal

, Saxena

, Katoch

CDS

. Transforming clinical virology with AI, machine learning and deep learning: A comprehensive review and outlook. Virusdisease 2023;34(3):345–355.

153.

Pagsuyoin

, Ng

, Molejon

, Luo

. Coupling wastewater-based epidemiology with data-driven machine learning for managing public health risks. Risk Anal 2025;45(10):2974–2982.

154.

Pang

, Xu

, Liu

, et al. The research progress of SARS-CoV-2 main protease inhibitors from 2020 to 2022. Eur J Med Chem 2023;257:115491.

155.

Patel

, Mall

, Bensmail

. AI-driven drug repurposing and binding pose meta dynamics identifies novel targets for monkeypox virus. J Infect Public Health 2023;16(5):799–807.

156.

Pinto

, Birdi

, Durant

, et al. Machine learning applications in population and public health: Guidelines for development, testing, and implementation. JMIR Public Health Surveill 2025;11:e68952.

157.

Popov

, Kalinin

, Buslaev

, et al. Unraveling viral drug targets: A deep learning-based approach for the identification of potential binding sites. Brief Bioinform 2023;25(1):bbad459.

158.

Puca

, Trent

. Using the Surveillance Tool EpiWATCH to Rapidly Detect Global Mumps Outbreaks. Global Biosecurity 2020;1(3).

159.

Pun

, Ozerov

, Zhavoronkov

. AI-powered therapeutic target discovery. Trends Pharmacol Sci 2023;44(9):561–572.

160.

Rajput

, Pramanik

, Nannaware

, et al. Wastewater surveillance in post-omicron silent phase uncovers silent waves and cryptic transmission of SARS-CoV-2 variants; a yearlong study in Western India. Sci Total Environ 2024;955:176833.

161.

Rehms

, Ellenbach

, Rehfuess

, et al. A Bayesian hierarchical approach to account for evidence and uncertainty in the modeling of infectious diseases: An application to COVID-19. Biom J 2024;66(1):e2200341.

162.

Reich

, McGowan

, Yamana

, et al. Accuracy of real-time multi-model ensemble forecasts for seasonal influenza in the U.S. PLoS Comput Biol 2019;15(11):e1007486.

163.

Roche

, Drake

, Rohani

. An Agent-Based Model to study the epidemiological and evolutionary dynamics of Influenza viruses. BMC Bioinformatics 2011;12:87.

164.

Rodríguez

, Kamarthi

, Agarwal

, et al. Machine learning for data-centric epidemic forecasting. Nat Mach Intell 2024;6(10):1122–1131.

165.

Sahai

, Gurukar

, KhudaBukhsh

, et al. A machine learning model for nowcasting epidemic incidence. Math Biosci 2022;343:108677.

166.

Saleem

, Al-Ghamdi

ASA

, Alassafi

, AlGhamdi

. Machine learning, deep learning, and mathematical models to analyze forecasting and epidemiology of COVID-19: A systematic literature review. Int J Environ Res Public Health 2022;19(9):5099.

167.

Satu

, Howlader

, Mahmud

, et al. Short-Term prediction of COVID-19 cases using machine learning models. Applied Sciences 2021;11(9):4266.

168.

Sawhney

RFB

, Dejean

, Schreiber

, et al. Fine-tuning protein language models unlocks the potential of underrepresented viral proteomes. PeerJ 2025;13:e19919.

169.

Sebastianelli

, Spiller

, Carmo

, et al. A reproducible ensemble machine learning approach to forecast dengue outbreaks. Sci Rep 2024;14(1):3807.

170.

Serrano

, Luciano

, Anaya

, et al. Artificial Intelligence (AI) applications in drug discovery and drug delivery: Revolutionizing personalized medicine. Pharmaceutics 2024;16(10):1328.

171.

Shafi

, Parwani

. Artificial intelligence in diagnostic pathology. Diagn Pathol 2023;18(1):109.

172.

Shah

SAW

, Palomar

, Barr

, et al. Seasonal antigenic prediction of influenza A H3N2 using machine learning. Nat Commun 2024;15(1):3833.

173.

Shah

, Ali

, Waqar

, Kim

. Federated learning in public health: A systematic review of decentralized, equitable, and secure disease prevention approaches. Healthcare (Basel) 2025;13(21):2760.

174.

Shahid

, Sattar

, Iqbal

, et al. Next-Generation Sequencing and the CRISPR-Cas nexus: A molecular plant virology perspective. Front Microbiol 2020;11:609376.

175.

Shi

, Huang

, Dang

, Zhou

. Leveraging social media data for pandemic detection and prediction. Humanit Soc Sci Commun 2024;11(1):1075.

176.

Shu

, McCauley

. GISAID: Global initiative on sharing all influenza data - from vision to reality. Euro Surveill 2017;22(13):30494.

177.

Sinno

, Baghdadi

, Narch

, et al. Charting the virosphere: Computational synergies of AI and bioinformatics in viral discovery and evolution. J Virol 2025;99(12):e01554–e1525.

178.

Sironi

, Kaderali

. Bioinformatics Algorithms and Predictive Models: The Grand Challenge in Computational Virology. FrontVirol 2021;1.

179.

Spreco

, Jöud

, Eriksson

, et al. Nowcasting (Short-Term Forecasting) of COVID-19 hospitalizations using syndromic healthcare data, Sweden, 2020. Emerg Infect Dis 2022;28(3):564–571.

180.

Steiner

, Gibson

, Crandall

. Drug resistance prediction using deep learning techniques on HIV-1 sequence data. Viruses 2020;12(5):560.

181.

Stone

, Gasparetto

, De Mattos

, et al. Wastewater monitoring as a tool to anticipate outbreaks and viral pandemics. The Brazilian Journal of Infectious Diseases 2026;30:105493.

182.

Sudhakar

, Bhansali

, Walkington

, Puelz

. The disutility of compartmental model forecasts during the COVID-19 pandemic. Front Epidemiol 2024;4:1389617.

183.

Sun

, Sun

, Yang

, et al. A novel, covalent broad-spectrum inhibitor targeting human coronavirus Mpro. Nat Commun 2025;16(1):4546.

184.

Suri

, Dakshanamurthy

. IntegralVac: A machine learning-based comprehensive multivalent epitope vaccine design method. Vaccines (Basel) 2022;10(10):1678.

185.

Thadani

, Gurev

, Notin

, et al. Learning from prepandemic data to forecast viral escape. Nature 2023;622(7984):818–825.

186.

Thomason

. Digital biology and the ethics of codified health systems. Global Health Journal 2026;10(1):1–6.

187.

Thompson

, Southall

, Daon

, et al. The impact of cross-reactive immunity on the emergence of SARS-CoV-2 variants. Front Immunol 2022;13:1049458.

188.

Tonks

, Harris

, Li

, et al. Forecasting West Nile Virus with graph neural networks: Harnessing spatial dependence in irregularly sampled geospatial data. Geohealth 2024;8(7):e2023GH000784.

189.

Tripathy

, Frohock

, Wang

, et al. Effective integration of multi-omics with prior knowledge to identify biomarkers via explainable graph neural networks. NPJ Syst Biol Appl 2025;11(1):43.

190.

Tulchinsky

, Zhao

, Kipshidze

, et al. Machine learning–driven COVID-19 hospitalization forecasting: From theory to practice in a major northeastern academic medical center. Open Forum Infect Dis 2025;12(6):ofaf307.

191.

Ullanat

, Jing

, Sledzieski

, Berger

. Learning the language of protein-protein interactions. Nat Commun 2026;17(1):1199.

192.

UNAIDS. (2025) Global HIV & AIDS statistics—Fact sheet. Joint United Nations Programme on HIV/AIDS (UNAIDS).

193.

UNESCO. (2021) Recommendation on the Ethics of Artificial Intelligence. United Nations Educational, Scientific and Cultural Organization. UNESCO: Paris, France.

194.

van der Horst

, Carter-Timofte

, Danneels

, et al. Large-scale deep learning identifies the antiviral potential of PKI-179 and MTI-31 against coronaviruses. Antiviral Res 2024;231:106012.

195.

Varadi

, Anyango

, Deshpande

, et al. AlphaFold protein structure database: Massively expanding the structural coverage of protein-sequence space with high-accuracy models. Nucleic Acids Res 2022;50(D1):D439–D444.

196.

Viboud

, Boëlle

, Carrat

, et al. Prediction of the spread of influenza epidemics by the method of analogues. Am J Epidemiol 2003;158(10):996–1006.

197.

Villanueva-Flores

, Sanchez-Villamil

, Garcia-Atutxa

. AI-driven epitope prediction: A systematic review, comparative analysis, and practical guide for vaccine development. NPJ Vaccines 2025;10(1):209.

198.

Wang

, Ye

, Liu

. Constructing big data prevention and control model for public health emergencies in China: A grounded theory study. Front Public Health 2023;11:1112547.

199.

Wang

, Luo

, Cai

, et al. DeepHVI: A multimodal deep learning framework for predicting human-virus protein-protein interactions using protein language models. Biosaf Health 2025;7(4):257–266.

200.

Watson

, Xiong

, Zhang

, et al. Pandemic velocity: Forecasting COVID-19 in the US with a machine learning & Bayesian time series compartmental model. PLoS Comput Biol 2021;17(3):e1008837.

201.

Watson

, Juergens

, Bennett

, et al. De novo design of protein structure and function with RFdiffusion. Nature 2023;620(7976):1089–1100.

202.

WHO. (2023) WHO Coronavirus (COVID-19) dashboard > More resources [Dashboard]. WHO.

203.

WHO. (2025a). WHO upgrades its public health intelligence system to boost global health security. ReliefWeb.

204.

WHO. (2025b). Epidemic Intelligence from Open Sources (EIOS) Strategy 2024–2026. WHO.

205.

Williams

, Zhan

C-G

. Staying ahead of the game: How SARS-CoV-2 has accelerated the application of machine learning in pandemic management. BioDrugs 2023;37(5):649–674.

206.

Wohlwend

, Nathan

, Shalon

, et al. Deep learning enhances the prediction of HLA class I-presented CD8+ T cell epitopes in foreign pathogens. Nat Mach Intell 2025;7(2):232–243.

207.

, Tanim

, Woo

, et al. A deep learning approach for enhancing pandemic prediction: A retrospective evaluation of transformer neural networks and multi-source data fusion for infectious disease forecasting. Epidemics 2025;53:100865.

208.

, Xu

, Zhang

, et al. Leveraging viral genome sequences and machine learning models for identification of potentially selective antiviral agents. Commun Chem 2025;8(1):189.

209.

Yan

, Lv

, Wen

, et al. PreTP-Stack: Prediction of therapeutic peptides based on the stacked ensemble learing. IEEE/ACM Trans Comput Biol Bioinform 2023;20(2):1337–1344.

210.

Yang

, Clifton

, Dung

, et al. Mitigating machine learning bias between high income and low–middle income countries for enhanced model fairness and generalizability. Sci Rep 2024;14(1):13318.

211.

Yang

, Li

, Yang

, et al. Deep-learning model for influenza prediction from multisource heterogeneous data in a megacity: Model development and evaluation. J Med Internet Res 2023;25:e44238.

212.

Yang

, Shan

, Cao

, et al. AI agents and epidemic intelligence on respiratory infectious diseases: Toward a conceptual framework integrating decision support. J Med Internet Res 2026;28:e86936.

213.

Yang

, Li

, Sun

, et al. Potential drug discovery for COVID-19 treatment targeting Cathepsin L using a deep learning-based strategy. Comput Struct Biotechnol J 2022;20:2442–2454.

214.

Yang

, Bogdan

, Nazarian

. An in silico deep learning approach to multi-epitope vaccine design: A SARS-CoV-2 case study. Sci Rep 2021;11(1):3238.

215.

Yao

, Zhang

, Li

, et al. A deep learning-based drug repurposing screening and validation for anti-SARS-CoV-2 compounds by targeting the cell entry mechanism. Biochem Biophys Res Commun 2023;675:113–121.

216.

, Pandey

, Bawden

, et al. Integrating artificial intelligence with mechanistic epidemiological modeling: A scoping review of opportunities and challenges. Nat Commun 2025;16(1):581.

217.

Yousaf

, Haseeb

, Shen

, et al. Data-driven discovery of antiviral peptides against PRRSV using multiple machine learning models. Front Vet Sci 2025;12:1681083.

218.

Zai

, Zhao

, Wang

, et al. Integrating protein language and geometric deep learning models for enhanced vaccine antigen prediction. Nat Commun 2025;17(1):1033.

219.

Zhang

, Zhai

, Bai

, et al. Towards fairness-aware and privacy-preserving enhanced collaborative learning for healthcare. Nat Commun 2025;16(1):2852.

220.

Zhao

, Zai

. Proceeding artificial intelligence technology towards vaccine designs. TIL 2025;3(4):100162.

221.

Zhao

, Song

. Antiviral Peptide-Generative Pre-Trained Transformer (AVP-GPT): A deep learning-powered model for antiviral peptide design with high-throughput discovery and exceptional potency. Viruses 2024;16(11):1673.

222.

Zhuang

, Vo

, Moshi

, et al. Early detection of emerging SARS-CoV-2 Variants from wastewater through genome sequencing and machine learning. Nat Commun 2025;16(1):6272.