Predicting Zoonotic Spillover: Computational Modeling of Receptor-Binding Dynamics in Emerging Bat Coronaviruses

By Dr. Zubair Khalid, DVM, MS, PhD · 2026-05-10

Key Takeaways

• Computational modeling, including homology modeling, molecular docking, and free energy calculations, is crucial for predicting the zoonotic potential of bat coronaviruses by analyzing the binding dynamics between the viral Receptor-Binding Domain (RBD) and host Angiotensin-Converting Enzyme 2 (ACE2).
• Sequence-based machine learning approaches, exemplified by the RAISE tool, integrate RBD sequence features and ACE2 similarity to generate spillover risk scores, enabling prioritization of viruses for surveillance and experimental validation.
• The variability of ACE2 receptor residues at the RBD binding interface (e.g., positions 31, 35, 38, and 353) significantly influences species susceptibility and is a key factor in predicting cross-species transmission events.
• Computational predictions directly inform veterinary surveillance by identifying bat coronaviruses with high affinity for livestock ACE2 receptors, guiding targeted serological and molecular monitoring in high-risk regions and animal populations.
• Understanding RBD-ACE2 interactions aids veterinary vaccine design by identifying conserved epitopes for eliciting broadly neutralizing antibodies, potentially protecting against multiple bat-origin coronaviruses with zoonotic potential.
• Challenges remain in accurately modeling highly divergent bat coronavirus RBDs and accounting for glycosylation and dynamic spike protein conformations, necessitating continued refinement of computational methods and integration with experimental data.

Abstract

Zoonotic spillover of bat coronaviruses into peridomestic and domestic animal species represents a critical pathway for the emergence of novel viral pathogens with pandemic potential. The receptor-binding domain (RBD) of the coronavirus spike protein and its interaction with host angiotensin-converting enzyme 2 (ACE2) is a primary molecular determinant of host range and cross-species transmission. This article reviews the computational methods used to predict RBD-ACE2 binding dynamics, including homology modeling, molecular docking, free energy calculations, and sequence-based machine learning. We discuss the integration of these approaches into surveillance frameworks such as the RAISE (Risk Assessment of Interspecies Spillover Events) tool and examine how structural predictions inform veterinary vaccine design and targeted wildlife monitoring. Emphasis is placed on bat-origin alphacoronaviruses and sarbecoviruses, with comparative analysis of ACE2 receptor variability across mammalian species. The article concludes with an assessment of current limitations and future directions for computational spillover risk modeling.

Introduction

Bat species of the order Chiroptera serve as reservoir hosts for a diverse array of coronaviruses, many of which have demonstrated the capacity for cross-species transmission to humans and domestic animals [<a href="#ref-1">1</a>, <a href="#ref-2">2</a>, <a href="#ref-3">3</a>]. Ecological and anthropogenic factors, including habitat encroachment and roosting in human structures, increase the frequency of bat-human and bat-livestock interfaces, thereby elevating spillover risk [<a href="#ref-4">4</a>, <a href="#ref-5">5</a>, <a href="#ref-6">6</a>]. The molecular gateway for coronavirus entry is the interaction between the viral spike protein RBD and host cell surface receptors, most commonly ACE2 for sarbecoviruses and related alphacoronaviruses [<a href="#ref-7">7</a>, <a href="#ref-8">8</a>]. Computational modeling of this interaction has become an indispensable tool for predicting which bat coronaviruses pose the greatest zoonotic threat, as experimental characterization of every novel virus is impractical [<a href="#ref-3">3</a>, <a href="#ref-9">9</a>].

The RBD-ACE2 binding interface is characterized by a complex network of hydrogen bonds, van der Waals contacts, and electrostatic interactions that determine binding affinity and specificity [<a href="#ref-8">8</a>]. Bat coronaviruses, particularly those from the subgenera Sarbecovirus and Merbecovirus, exhibit RBD sequences that can be phylogenetically grouped and structurally modeled to infer receptor compatibility [<a href="#ref-10">10</a>, <a href="#ref-11">11</a>]. Computational structural biology methods, including homology modeling, molecular docking, and molecular dynamics (MD) simulations, allow researchers to construct three-dimensional models of RBD-ACE2 complexes and compute binding free energies [<a href="#ref-8">8</a>, <a href="#ref-12">12</a>]. These predictions are then integrated with epidemiological data, viral genomics, and machine learning classifiers to produce spillover risk scores [<a href="#ref-9">9</a>, <a href="#ref-11">11</a>, <a href="#ref-13">13</a>].

This article focuses exclusively on the computational modeling of receptor-binding dynamics in bat coronaviruses relevant to veterinary medicine and wildlife surveillance. Human clinical outcomes are discussed only where direct comparative host-range parallels are drawn, such as the use of human ACE2 as a reference receptor for evaluating zoonotic potential in animal models [<a href="#ref-12">12</a>, <a href="#ref-14">14</a>].

Computational Methods for Predicting RBD-ACE2 Interactions

Homology Modeling of Bat Coronavirus RBDs

The three-dimensional structure of the RBD is not experimentally determined for every bat coronavirus isolate. Homology modeling, also known as comparative modeling, provides a reliable alternative when a template structure with sufficient sequence identity exists [<a href="#ref-15">15</a>]. The RBD core structure is conserved among coronaviruses, allowing the use of solved structures such as SARS-CoV-2 RBD (PDB 6M0J) or SARS-CoV RBD (PDB 2AJF) as templates for bat sarbecovirus RBDs [<a href="#ref-12">12</a>]. Tools such as MODELLER and SWISS-MODEL are commonly employed to generate models based on sequence alignment [<a href="#ref-15">15</a>]. The quality of the model is evaluated using DOPE scores, QMEAN, and Ramachandran plots to ensure stereochemical plausibility [<a href="#ref-15">15</a>]. For bat alphacoronaviruses that use receptors other than ACE2, such as CEACAM6, homology modeling is adapted to the specific receptor-binding region [<a href="#ref-16">16</a>].

Molecular Docking of RBD-ACE2 Complexes

Once a structural model of the bat coronavirus RBD is built, molecular docking is used to predict the binding pose and affinity with host ACE2 orthologs [<a href="#ref-12">12</a>, <a href="#ref-15">15</a>]. Rigid-body docking algorithms, such as those implemented in ClusPro or ZDOCK, evaluate millions of potential orientations and score them based on shape complementarity and electrostatic compatibility [<a href="#ref-15">15</a>]. Flexible docking approaches, which allow side-chain conformational changes, are often necessary to capture induced-fit effects at the interface [<a href="#ref-12">12</a>]. The docking score, typically expressed as a binding energy or a probability score, provides a quantitative estimate of the likelihood of receptor recognition [<a href="#ref-12">12</a>, <a href="#ref-15">15</a>]. For example, screening candidate intermediate hosts for porcine respiratory coronavirus using molecular docking revealed species-specific binding preferences that correlated with known susceptibility [<a href="#ref-15">15</a>].

Free Energy Calculations and Molecular Dynamics Simulations

Molecular dynamics (MD) simulations offer a more detailed and dynamic view of the RBD-ACE2 interaction than static docking. By simulating the solvated complex over nanosecond to microsecond timescales, MD captures conformational fluctuations, water-mediated hydrogen bonds, and the stability of the binding interface [<a href="#ref-8">8</a>, <a href="#ref-12">12</a>]. Free energy perturbation (FEP) and molecular mechanics generalized Born surface area (MM/GBSA) methods are then applied to compute the binding free energy (ΔG) of the complex [<a href="#ref-12">12</a>]. These calculations can distinguish between high-affinity and low-affinity interactions and identify specific residues that contribute disproportionately to binding [<a href="#ref-12">12</a>]. For instance, MM/GBSA analysis of bat sarbecovirus RBDs with human ACE2 has revealed that mutations at residue positions 493, 498, and 501 can significantly alter binding affinity, thereby modulating spillover potential [<a href="#ref-12">12</a>, <a href="#ref-17">17</a>].

Sequence-Based Machine Learning for Spillover Risk Prediction

Machine learning models trained on viral genomic and proteomic features have been developed to predict zoonotic potential without requiring a solved structure [<a href="#ref-9">9</a>, <a href="#ref-13">13</a>, <a href="#ref-18">18</a>]. Features such as RBD sequence composition, predicted secondary structure, and phylogenetic distance to known zoonotic viruses are used as inputs for classifiers like random forest, support vector machines, and deep neural networks [<a href="#ref-13">13</a>, <a href="#ref-18">18</a>, <a href="#ref-19">19</a>]. The RAISE (Risk Assessment of Interspecies Spillover Events) tool exemplifies this approach, integrating RBD sequence features with ACE2 sequence similarity across vertebrate hosts to produce a spillover risk score for sarbecoviruses [<a href="#ref-11">11</a>]. Other models incorporate host ecological traits, such as anthropogenic roosting behavior, to predict which bat species are most likely to harbor viruses with high zoonotic potential [<a href="#ref-4">4</a>, <a href="#ref-5">5</a>, <a href="#ref-20">20</a>]. A critical challenge in machine learning for spillover prediction is the limited availability of experimentally validated negative examples (i.e., viruses that do not infect humans), which can lead to overestimation of risk [<a href="#ref-9">9</a>, <a href="#ref-18">18</a>].

Case Studies: Computational Modeling of Bat Coronavirus RBDs

Sarbecovirus RBD-ACE2 Binding and the RAISE Framework

Sarbecoviruses, including SARS-CoV, SARS-CoV-2, and numerous bat isolates, use ACE2 as the primary entry receptor. The RAISE computational tool was developed to systematically evaluate the spillover potential of bat sarbecoviruses by combining RBD-ACE2 docking scores, sequence conservation of critical binding residues, and phylogenetic relatedness to known human-infecting viruses [<a href="#ref-11">11</a>]. In a large-scale screening, RAISE identified several bat sarbecoviruses with predicted high binding affinity to human ACE2, many of which were subsequently confirmed in vitro [<a href="#ref-11">11</a>]. The tool also assesses the likelihood of spillover into intermediate hosts, such as pangolins and civets, by docking bat RBDs against ACE2 orthologs from a panel of mammalian species [<a href="#ref-11">11</a>, <a href="#ref-14">14</a>]. Comparative analysis of ACE2 sequences from bats, rodents, and carnivores reveals that certain residues at the RBD binding interface (e.g., positions 31, 35, 38, and 353) are highly variable and can determine species susceptibility [<a href="#ref-12">12</a>, <a href="#ref-14">14</a>].

Alphacoronavirus RBD Interactions with Alternative Receptors

Not all bat coronaviruses use ACE2. Heart-nosed bat alphacoronaviruses have been shown to use human CEACAM6 as an entry receptor, a finding that emerged from a combination of homology modeling and viral pseudotype entry assays [<a href="#ref-16">16</a>]. Computational docking of the alphacoronavirus RBD to CEACAM6 revealed a binding interface distinct from that of sarbecovirus-ACE2 complexes, highlighting the need for receptor-specific modeling approaches [<a href="#ref-16">16</a>]. Similarly, porcine respiratory coronavirus uses aminopeptidase N (APN) as a receptor, and molecular docking studies have been used to screen potential intermediate hosts based on APN sequence variation [<a href="#ref-15">15</a>]. These examples underscore the importance of expanding computational models to include non-ACE2 receptors, as many bat coronaviruses may use alternative molecules for cell entry.

Implications for Veterinary Surveillance and Vaccine Design

Computational predictions of RBD-ACE2 binding dynamics directly inform surveillance strategies for emerging bat coronaviruses in domestic animal populations. By identifying bat coronaviruses with high predicted affinity for livestock ACE2 receptors, targeted serological and molecular surveillance can be implemented in high-risk regions [<a href="#ref-13">13</a>, <a href="#ref-21">21</a>]. For example, models predicting that bat sarbecoviruses can bind to ACE2 of pigs, cattle, and dogs have prompted surveillance in Southeast Asian pig farms and live animal markets [<a href="#ref-12">12</a>, <a href="#ref-21">21</a>]. The RAISE tool provides a prioritized list of bat coronaviruses for which intermediate host screening should be conducted [<a href="#ref-11">11</a>].

In veterinary vaccine design, knowledge of the RBD-ACE2 interface enables the selection of immunogens that elicit broadly neutralizing antibodies across multiple sarbecovirus lineages. Structural modeling of the RBD bound to ACE2 can identify conserved epitopes that are less prone to escape mutations [<a href="#ref-22">22</a>]. A trivalent vaccine strategy based on immunogenic relationship mapping of sarbecovirus RBDs has been proposed to provide broad protection against bat-origin coronaviruses with zoonotic potential [<a href="#ref-22">22</a>]. Computational design of RBD-based subunit vaccines relies on accurate modeling of the RBD architecture to ensure proper folding and antigen presentation [<a href="#ref-22">22</a>].

The following Mermaid diagram illustrates the typical computational workflow for assessing spillover risk from bat coronavirus RBD sequences.

flowchart TD
 A["Bat coronavirus RBD sequence"] --> B["Homology modeling of RBD 3D structure"]
 B --> C["Selection of host ACE2 orthologs"]
 C --> D["Molecular docking of RBD-ACE2 complex"]
 D --> E["Free energy calculation (MM/GBSA or FEP)"]
 E --> F{"Binding affinity threshold exceeded?"}
 F -->|"Yes"| G["High spillover risk candidate"]
 F -->|"No"| H["Low risk"]
 G --> I["In vitro binding validation"]
 I --> J["Field surveillance in intermediate hosts"]
 G --> K["Machine learning risk classifier (e.g., RAISE)"]
 K --> L["Spillover risk score"]
 L --> M["Surveillance prioritization"]

Challenges and Future Directions

Despite significant advances, computational modeling of receptor-binding dynamics faces several limitations. First, the accuracy of homology models depends on template availability; RBDs from highly divergent bat coronaviruses may not be modeled reliably [<a href="#ref-15">15</a>]. Second, docking and free energy calculations often ignore the role of glycosylation on both the RBD and the receptor, which can affect binding kinetics [<a href="#ref-8">8</a>, <a href="#ref-12">12</a>]. Third, machine learning models are constrained by data imbalance and the difficulty of obtaining true negative labels for non-zoonotic viruses [<a href="#ref-9">9</a>, <a href="#ref-18">18</a>]. Fourth, the dynamic nature of the spike protein, including conformational changes between closed and open states, requires enhanced sampling MD simulations that are computationally expensive [<a href="#ref-12">12</a>].

Future directions include the integration of deep learning models, such as AlphaFold2, for predicting RBD structures with high accuracy even in the absence of close templates [<a href="#ref-11">11</a>]. The application of large language models trained on viral nucleotide sequences has shown promise in improving the generalizability of spillover predictions across viral families [<a href="#ref-18">18</a>]. Additionally, combining in silico predictions with in vitro binding assays and ecological data from bat roosting ecology will enhance the resolution of spillover risk maps [<a href="#ref-4">4</a>, <a href="#ref-5">5</a>, <a href="#ref-20">20</a>]. The table below summarizes the main computational methods discussed and their applications.

Conclusion

Computational modeling of receptor-binding dynamics provides a powerful framework for predicting zoonotic spillover potential of emerging bat coronaviruses. Homology modeling, molecular docking, free energy calculations, and machine learning each contribute unique insights into the molecular determinants of host range. The RAISE tool and similar platforms now allow systematic triage of bat coronaviruses for surveillance and vaccine design. Continued refinement of these methods, coupled with experimental validation and ecological data, will be essential for preempting the next zoonotic coronavirus spillover event into domestic animal populations.

References

Related Clinical & Scientific Guides

• A Practical Guide to Detecting Antimicrobial Resistance Genes in Shotgun Metagenomic Data
• Computational Immunology: Modeling the Immune System
• How to Set Hard Filters for Germline Variant Calling: A Practical Guide to GATK Best Practices