African Swine Fever: Computational Models for Early Detection and Spread Prediction in Wild Boar Populations

Abstract

African swine fever (ASF) remains a catastrophic threat to global swine production with wild boar (Sus scrofa) serving as the primary maintenance host in Europe and parts of Asia. This review synthesizes current computational approaches for early detection and spread prediction, integrating agent-based modeling (ABM), patch-based stochastic frameworks, environmental sampling via quantitative polymerase chain reaction (qPCR), and geographic information system (GIS) mapping. We examine the biophysical mechanisms driving viral persistence in wild boar populations, the algorithmic architectures underpinning predictive models, and the operational integration of molecular diagnostics with geospatial analytics. Particular emphasis is placed on European Union outbreak data, zonal control strategies, and biosecurity barrier efficacy assessment through dynamical modeling approaches.

1. Introduction

African swine fever virus (ASFV), a large double-stranded DNA virus belonging to the Asfarviridae family, exhibits exceptional environmental stability and host-specific pathogenicity in suids. The current genotype II pandemic, originating in Georgia in 2007, has established endemic cycles in wild boar populations across Eastern and Central Europe with sporadic incursions into domestic pig holdings. The epidemiological triad of ASFV persistence comprises: (i) direct transmission between infected and susceptible wild boar, (ii) indirect transmission via contaminated environment and carcasses, and (iii) human-mediated long-distance jumps through contaminated fomites, vehicles, and pork products.

Wild boar population dynamics, characterized by high reproductive rates, social structuring into sounders, and seasonal movement patterns, create complex spatiotemporal transmission networks that defy simple compartmental modeling. Computational approaches must therefore incorporate individual heterogeneity, landscape connectivity, and environmental viral decay kinetics to generate actionable predictions for veterinary authorities.

2. Viral Pathogenesis and Host-Virus Interactions in Wild Boar

2.1 Molecular Determinants of Virulence and Immune Evasion

ASFV encodes over 150 proteins, many of which modulate host innate and adaptive immune responses. The viral protein pA137R triggers inflammatory responses by inducing the NF-κB signaling pathway and facilitating NLRP3 inflammasome assembly, contributing to the characteristic hemorrhagic pathology [8]. Concurrently, the B66L protein drives extracellular mitochondrial release to promote systemic inflammation, exacerbating tissue damage and viral dissemination [12]. These mechanisms underlie the acute lethal phenotype observed in European wild boar, where case fatality rates exceed 90 percent for genotype II isolates.

2.2 Host Genetic Resistance and Immunological Correlates

Integrated multi-omics profiling has identified genetic loci associated with ASF resistance in domestic pigs, revealing candidate genes involved in interferon signaling, macrophage activation, and cytotoxic lymphocyte function [9]. In wild boar, distinct cytotoxic cell subsets underlie protective versus non-protective immunity, with specific natural killer (NK) cell and CD8+ T-cell phenotypes correlating with survival outcomes [1]. These findings inform parameterization of host susceptibility distributions in computational models, enabling simulation of heterogeneous infection outcomes rather than binary susceptible-infected transitions.

2.3 In Vitro Systems for Viral Kinetics Parameterization

Cryopreserved primary swine macrophages serve as physiologically relevant substrates for ASFV replication kinetics studies, providing quantitative data on viral growth curves, multiplicity of infection dynamics, and cell-to-cell transmission efficiency [14]. These in vitro systems generate the kinetic parameters, eclipse phase duration, burst size, infectious virion half-life, required for within-host submodels embedded within population-level transmission frameworks.

3. Computational Modeling Frameworks

3.1 Agent-Based Modeling Architecture

Agent-based models (ABMs) represent individual wild boar as autonomous agents with state variables including: infection status (susceptible, exposed, infectious, recovered, deceased), age, sex, sounder affiliation, spatial coordinates, and immunological profile. The core algorithmic loop operates on discrete time steps (typically daily resolution) with the following processes:

Movement Submodel: Agents execute correlated random walks modulated by habitat suitability indices derived from land cover data, seasonal resource availability, and anthropogenic barriers. Sounders maintain cohesion through attraction-repulsion kernels with distance-dependent interaction radii.

Transmission Submodel: Direct transmission occurs when infectious and susceptible agents occupy proximal spatial cells (typically <50 meters) with probability β_direct modulated by contact duration and viral load. Indirect transmission via environmental contamination implements a spatially explicit viral load field where carcasses and excretions deposit infectious virions that decay according to temperature-dependent kinetics.

Demographic Submodel: Reproduction follows seasonal breeding patterns with litter sizes drawn from empirical distributions. Natural mortality incorporates age-specific hazards. Disease-induced mortality removes infectious agents after a gamma-distributed infectious period.

3.2 Patch-Based Stochastic Frameworks

Patch-based models discretize the landscape into epidemiological units (patches) representing management zones, hunting districts, or habitat fragments. A patch-based stochastic framework applied to the Republic of Korea demonstrated the utility of metapopulation approaches for national-scale risk assessment [15]. The master equation for patch i tracks the probability distribution of infected individuals:

dP(I_i,t)/dt = Σ_j [β_ij S_i I_j P(·)] - γ I_i P(·) + μ (I_i+1) P(·) - μ I_i P(·)

where β_ij represents the transmission kernel between patches i and j incorporating wild boar dispersal probabilities, γ is the recovery/removal rate, and μ represents demographic turnover. Stochastic simulation via Gillespie algorithm captures extinction-recolonization dynamics critical for low-prevalence scenarios.

3.3 Dynamical Modeling of Zonal Control Strategies

Dynamical modeling of zonal prevention and control under normalized management has been applied to ASF transmission in China, demonstrating the impact of movement restrictions, depopulation radii, and surveillance intensity on epidemic trajectories [11]. The model structure incorporates:

• Core zone: Infected area with movement ban and enhanced carcass removal
• Buffer zone: Surrounding area with intensified surveillance and hunting restrictions
• Surveillance zone: Outer perimeter with passive surveillance and biosecurity enforcement

The system of delay differential equations captures the time lag between infection occurrence, detection, and intervention implementation:

dS/dt = Λ - β S(t) I(t-τ) - μ S
dE/dt = β S(t) I(t-τ) - (σ + μ) E
dI/dt = σ E - (γ + μ + α) I
dR/dt = γ I - μ R

where τ represents the detection delay, α is disease-induced mortality, and Λ represents recruitment. Sensitivity analysis identifies detection delay τ as the dominant parameter controlling epidemic size, emphasizing the premium on early detection systems.

4. Environmental Surveillance and Molecular Diagnostics

4.1 Environmental Sampling via qPCR

Environmental sampling targets ASFV DNA in wild boar carcasses, feces, soil, water, and vegetation. Quantitative PCR targeting the B646L (p72) gene provides standardized viral load quantification with detection limits of 10-100 genome copies per reaction. The molecular identification and phylogenetic profiling of ASFV in Indonesia during 2021-2025 based on the B646L gene demonstrates the utility of this target for both diagnostics and molecular epidemiology [2].

Sampling Design: Systematic grid-based sampling (2×2 km cells) with adaptive allocation to high-risk habitats (wetlands, oak forests, agricultural margins). Carcass sampling prioritizes spleen, lymph nodes, and bone marrow where viral loads peak. Environmental swabs from feeding sites and wallows capture shed virus.

qPCR Workflow:

1. Sample collection in viral transport medium with RNase/DNase inhibitors
2. Automated nucleic acid extraction using magnetic bead-based protocols
3. Real-time PCR with hydrolysis probe chemistry (FAM-labeled probe, BHQ1 quencher)
4. Quantification via standard curve of plasmid-cloned B646L target
5. Cycle threshold (Ct) values converted to genome copies per gram/mL using extraction efficiency controls

Data Integration: Georeferenced qPCR results feed directly into GIS layers for spatial risk mapping and model calibration. Positive environmental samples trigger targeted carcass searches and hunting bag surveillance in surrounding grid cells.

4.2 Rapid Antigen Detection for Field Deployment

A rapid detection method for ASFV antigens in serum and whole blood samples using an automated machine reading tool for lateral flow assays enables pen-side diagnostics with 15-minute turnaround [5]. The assay targets the p72 and p30 structural proteins with monoclonal antibody pairs. Automated readers eliminate subjective interpretation and provide quantitative signal intensity correlating with viral load. Field validation demonstrates sensitivity of 85-92 percent and specificity >98 percent relative to qPCR in acute infection phases. Integration with mobile data capture platforms enables real-time upload to central databases for model assimilation.

4.3 Serological Surveillance and Antibody Kinetics

Enzyme-linked immunosorbent assay (ELISA) targeting the p30, p54, and p72 proteins detects antibodies from 7-10 days post-infection. Seroprevalence surveys in wild boar hunting bags provide retrospective infection pressure estimates. However, the acute lethal nature of genotype II infection limits seropositive detection; most infected wild boar die before seroconversion. Serology therefore primarily identifies rare survivors and provides evidence of historical circulation in endemic zones.

5. Geospatial Analysis and GIS Mapping

5.1 Landscape Epidemiology and Habitat Suitability

GIS mapping integrates multiple data layers to generate dynamic risk surfaces:

Habitat Suitability Modeling: Maximum entropy (MaxEnt) and ensemble species distribution models combine presence-only wild boar occurrence data with environmental predictors (forest cover, mast production, water distance, snow depth, human disturbance index) to generate continuous suitability surfaces. These surfaces weight agent movement probabilities and carrying capacities in ABMs.

5.2 Spatiotemporal Cluster Detection

Space-time permutation scan statistics (SaTScan) identify emerging clusters of ASFV-positive wild boar. The cylindrical scanning window varies in spatial radius (0-50 km) and temporal height (0-90 days) to detect clusters without assuming predefined risk areas. Cluster signals trigger adaptive surveillance intensification and model re-parameterization.

5.3 Connectivity Networks and Dispersal Corridors

Circuit theory and least-cost path analysis quantify landscape connectivity between wild boar subpopulations. Resistance surfaces assign movement costs to land cover types (forest=1, agriculture=5, urban=50, highway=100). Current flow maps identify pinch points where targeted barriers or intensified surveillance yield disproportionate risk reduction. These networks parameterize the β_ij transmission kernel in patch-based models.

6. Biosecurity Barriers and Zonal Control Strategies

6.1 Physical Barrier Efficacy Assessment

Physical barriers, fencing, highway mitigation structures, river corridors, modify wild boar movement kernels and reduce cross-boundary transmission. Computational experiments in ABMs simulate barrier scenarios by imposing infinite resistance on selected edges in the movement graph. Key findings:

• Continuous fencing >1.5 m height with buried apron reduces cross-fence movement by 85-95 percent
• Highway fencing combined with wildlife overpasses at 5 km intervals maintains population connectivity while reducing roadkill and long-distance dispersal
• River corridors act as semi-permeable barriers; seasonal ice cover increases winter crossing rates 3-5 fold

6.2 Biosecurity Risks in Smallholder Systems

Biosecurity risks for ASF occurrence in smallholder systems have been characterized in Gauteng, South Africa, identifying critical control points: uncontrolled swill feeding, inadequate perimeter fencing, shared equipment, and informal pig trade networks [3]. These findings translate to European contexts where backyard holdings interface with wild boar habitat. Computational models incorporate holding-level biosecurity scores as modifiers of the domestic-wild interface transmission coefficient.

6.3 Manure Management and Environmental Decontamination

Manure management during ASF outbreaks presents critical gaps in environmental decontamination protocols [10]. ASFV persists in slurry for weeks at 4°C and days at 20°C. Modeling of manure treatment pathways (anaerobic digestion, composting, lime treatment) quantifies residual infectivity and informs movement restrictions for organic fertilizers from infected zones. Thermal inactivation kinetics follow first-order decay with D-values of 15-30 minutes at 60°C depending on matrix composition.

6.4 Operational Response Integration

Firefighter involvement in animal disease response in Poland demonstrates the role of civil protection assets in carcass removal and disinfection operations [13]. Nationwide descriptive analysis reveals that rapid deployment of specialized teams reduces environmental viral load by 60-70 percent within the critical first 72 hours post-detection. Computational models incorporate carcass removal rates as time-dependent functions of response capacity.

7. Multi-Omics and Host Resistance Genetics

7.1 Resistance Loci and Selective Breeding Prospects

Integrated multi-omics profiling identifies genetic loci of ASF resistance in pigs, highlighting variants in TLR3, IRF7, STAT1, and GBP5 genes associated with reduced viral replication and moderated inflammatory response [9]. While wild boar populations cannot be selectively bred, these markers enable genomic surveillance of resistance allele frequencies in endemic zones. Computational models can simulate evolutionary trajectories of host resistance under sustained selection pressure.

7.2 Vaccine Development and Immunoinformatics

Advances in ASF vaccine development face challenges including lack of neutralizing antibody correlates, viral immune evasion mechanisms, and safety concerns with live attenuated candidates [6]. In silico pharmacological analysis of Tinospora cordifolia compounds targeting ASFV B175L illustrates the application of structure-based virtual screening for antiviral discovery [4]. Identification of efficient multi-epitope combinations against ASFV based on AP205 scaffold-mediated nanodisplay technology represents a promising subunit vaccine platform [7]. Computational immunoinformatics pipelines predict MHC class I/II binding affinities, population coverage, and epitope conservation across ASFV genotypes to prioritize vaccine candidates.

8. Operational Integration and Decision Support

8.1 Model-Assimilated Surveillance Data

The computational modeling workflow integrates multiple data streams through a Bayesian model updating framework:

flowchart TD
    A[Environmental qPCR Results], > D[Data Assimilation Engine]
    B[Carcass Detection Reports], > D
    C[Hunting Bag Serology/PCR], > D
    D, > E[Posterior Parameter Distributions]
    E, > F[Agent-Based Model Ensemble]
    F, > G[Spatiotemporal Risk Forecasts]
    G, > H[Decision Support Dashboard]
    H, > I[Adaptive Surveillance Allocation]
    H, > J[Barrier Deployment Optimization]
    H, > K[Carcass Removal Prioritization]
    I, > A
    J, > B
    K, > C

8.2 Ensemble Forecasting and Uncertainty Quantification

Ensemble modeling combines ABM, patch-based stochastic, and dynamical model outputs through weighted averaging where weights derive from historical predictive performance (continuous ranked probability score). Prediction intervals capture parametric uncertainty (via Latin hypercube sampling), structural uncertainty (via multi-model ensemble), and scenario uncertainty (via alternative management interventions).

8.3 Decision-Theoretic Resource Allocation

Resource allocation optimization frames surveillance and control as a partially observable Markov decision process (POMDP). The state space includes true (unobserved) infection status across grid cells. Actions include: targeted carcass search, hunter-based sampling, barrier construction, depopulation. The reward function balances surveillance cost, control cost, and expected damage from undetected spread. Approximate dynamic programming yields near-optimal policies for real-time decision support.

9. Model Validation and Performance Metrics

9.1 Retrospective Validation Against EU Outbreak Data

Models are validated against the European Food Safety Authority (EFSA) ASF database comprising >20,000 wild boar cases across 15 member states since 2014. Validation metrics include:

• Spatial accuracy: Area under the receiver operating characteristic curve (AUC-ROC) for 30-day ahead risk maps (typical range 0.75-0.85)
• Temporal accuracy: Mean absolute error in weekly case incidence (typical range 15-30 percent)
• Cluster detection: Sensitivity/specificity of SaTScan alerts (sensitivity 0.70-0.80 at 90-day lead time)
• Barrier impact: Counterfactual comparison of observed vs. simulated spread with/without fencing

9.2 Prospective Validation in Disease-Free Regions

Prospective validation in disease-free regions (e.g., Scandinavia, Western Europe) uses simulated incursions seeded at high-risk entry points (border crossings, landfills, military zones). Models predict time-to-detection, outbreak size, and probability of establishment under current surveillance regimes. These exercises inform preparedness planning and surveillance gap analysis.

10. Computational Infrastructure and Reproducibility

10.1 High-Performance Computing Requirements

ABM ensembles with 10⁴-10⁵ agents over 10⁴ km² landscapes at daily resolution for 365-day simulations require 50-200 core-hours per replicate. Cloud-based HPC platforms (SLURM, Kubernetes) enable parallel execution of 100-500 ensemble members. Containerization (Docker/Singularity) ensures reproducibility across compute environments.

10.2 Data Standards and Interoperability

The Animal Disease Notification System (ADNS) and WOAH WAHIS databases provide standardized case reporting. Environmental sample metadata follow the Minimum Information about a Quantitative PCR Experiment (MIQE) guidelines. Geospatial data adhere to OGC standards (WMS, WFS, GeoPackage). Model parameters and code are version-controlled (Git) with persistent identifiers (DOI) via institutional repositories.

11. Emerging Directions

11.1 Integration of Genomic Surveillance

Phylogenomic analysis of ASFV B646L and whole-genome sequences enables reconstruction of transmission networks at fine spatiotemporal scales. Molecular clock models estimate time of most recent common ancestor (tMRCA) for outbreak clusters. Integration of genomic distance matrices into transmission kernels (β_ij ∝ exp(-α × genetic_distance)) improves inference of cryptic transmission chains.

11.2 Machine Learning Hybridization

Machine learning algorithms for predicting veterinary viral outbreaks complement mechanistic models by learning complex nonlinear relationships from high-dimensional surveillance data [16]. Gradient boosting machines and graph neural networks trained on historical ASF data predict weekly case counts at 10 km resolution with improved short-term accuracy (1-4 weeks) compared to mechanistic models alone. Hybrid approaches use ML for short-term nowcasting and mechanistic models for long-term scenario projection.

11.3 One Health and Cross-Species Surveillance

While ASFV does not infect humans, the computational infrastructure developed for wild boar surveillance is transferable to other wildlife-livestock interface pathogens including classical swine fever, porcine reproductive and respiratory syndrome, and zoonotic agents such as hepatitis E virus. Shared data pipelines, modeling frameworks, and decision support tools maximize return on investment in veterinary computational infrastructure.

12. Conclusions

Computational models for ASF early detection and spread prediction in wild boar populations have matured from theoretical constructs to operational decision support tools deployed by veterinary authorities across Europe and Asia. The integration of agent-based modeling, patch-based stochastic frameworks, environmental qPCR surveillance, GIS mapping, and dynamical systems analysis provides a multi-scale perspective spanning individual host-pathogen interactions to continental epidemic dynamics. Critical success factors include: (i) high-resolution wild boar density and movement data, (ii) rapid environmental diagnostic turnaround, (iii) adaptive model updating with surveillance data, and (iv) decision-theoretic translation of forecasts into resource allocation. Continued investment in genomic surveillance integration, machine learning hybridization, and cross-pathogen platform interoperability will enhance preparedness for ASF and emerging swine pathogens.

References

[1] Marín-Moraleda D, Tort-Miró A, Ezcurra E et al. Distinct cytotoxic cell subsets underlie protective and non-protective immunity to African swine fever virus. Emerg Microbes Infect. 2026. https://pubmed.ncbi.nlm.nih.gov/42246494/

[2] Saepulloh M, Dharmayanti NLPI, Khairullah AR et al. Molecular identification and phylogenetic profiling of African swine fever virus in Indonesia during 2021-2025 based on the B646L (p72) gene. Vet World. 2026. https://pubmed.ncbi.nlm.nih.gov/42245454/

[3] Montsu KP, Boshoff CI, Etter E et al. Biosecurity risks for African swine fever occurrence in smallholder systems in Gauteng, South Africa. Onderstepoort J Vet Res. 2026. https://pubmed.ncbi.nlm.nih.gov/42240317/

[4] Karunarathne K, Poorni S, Jayampathi N et al. In silico pharmacological analysis of Tinospora cordifolia compounds targeting African swine fever virus B175L. PLoS One. 2026. https://pubmed.ncbi.nlm.nih.gov/42234686/

[5] Olesen AS, Valadkhani B, Johnston CM et al. A rapid detection method for African swine fever virus antigens in serum and whole blood samples using an automated machine reading tool for lateral flow assays. J Virol Methods. 2026. https://pubmed.ncbi.nlm.nih.gov/42229692/

[6] Utomo DIS, Dharmayanti NI, Putra MY. Advances in African swine fever vaccine development: challenges and prospects. J Vet Sci. 2026. https://pubmed.ncbi.nlm.nih.gov/42225599/

[7] Lu H, Shao J, Liu W et al. Identification of efficient multi-epitope combinations against African swine fever virus based on AP205 scaffold-mediated nanodisplay technology. J Nanobiotechnology. 2026. https://pubmed.ncbi.nlm.nih.gov/42218507/

[8] Zeng Y, Liu X, Dong S et al. ASFV pA137R protein triggers inflammatory response by inducing NF-κB signaling pathway and facilitating NLRP3 inflammasome assembly. Virol Sin. 2026. https://pubmed.ncbi.nlm.nih.gov/42217584/

[9] Ye X, Xie Q, Cao C et al. Integrated multi-omics profiling identifies genetic loci of African swine fever resistance in pigs. Gigascience. 2026. https://pubmed.ncbi.nlm.nih.gov/42216858/

[10] Macdonald A, Omoaka N, Juarez M et al. Manure management during African swine fever outbreak: identifying gaps through a narrative review. Front Vet Sci. 2026. https://pubmed.ncbi.nlm.nih.gov/42211335/

[11] Li J, Zhang J, Gao L et al. Dynamical modeling and analysis of the impact of zonal prevention and control under normalized management on African Swine Fever transmission in China. Infect Dis Model. 2026. https://pubmed.ncbi.nlm.nih.gov/42205148/

[12] Lin L, Hu Z, Xing G et al. African swine fever virus B66L drives extracellular mitochondrial release to promote systemic inflammation in mice. Nat Commun. 202