Computational simulations reveal critical insights into the dynamics of gene drive, illuminating pathways toward safe and responsible genetic technologies. Discover how.

This version (Version 4) was created on 17 November 2025

Imagine a small island in the tropics, where malaria has plagued communities for generations. Authorities, equipped with gene drive technology, release thousands of modified mosquitoes into the wild. The goal? To stop malaria transmission in its tracks. But what if something unexpected happens? What if the mosquitoes spread beyond the island, or the intervention has unwanted effects on other species? These uncertainties highlight why predicting the behaviour of such interventions before their release is critical.

1. The need for predictive modelling in gene drive research

The release of engineered gene drive (EGD) organisms into the environment has yet to occur, creating significant uncertainties for regulators and researchers alike (Frieß et al., 2023). Traditional risk assessment methods used for genetically modified organisms rely on controlled experiments (many in the environment) and comparisons with known products, enabling regulators to estimate risks with relative confidence (Golnar 2021). But gene drives are different.

Unlike other genetic modifications, EGDs can spread through populations rapidly, potentially persisting for generations and affecting entire ecosystems in ways we may not yet fully understand nor want (Kim et al., 2023). Their risks cannot be evaluated solely through short-term, small-scale experiments. For instance, suppression drives, designed to reduce population sizes by targeting reproductive genes, could potentially have cascading effects on predator-prey dynamics and food availability (Golnar 2021). To bridge the gap between laboratory findings and complex ecological realities, predictive modelling plays an essential role.

By simulating different scenarios and ecological conditions, models help researchers estimate the likelihood of success, identify potential risks, and refine strategies before any real-world intervention (Rode et al., 2019). These simulations guide decision-making and ensure that interventions are deployed responsibly (Golnar et al., 2021).

A structured approach known as the phased testing pathway (Figure 1) evaluates gene drives step-by-step, progressing from controlled environments to field trials, with modelling playing a key role at every stage (Combs et al., 2023). By integrating ecological, genetic, and epidemiological data, models provide a comprehensive picture of possible outcomes and inform policy decisions.

Gene drive systems will be influenced by numerous ecological and evolutionary factors, making it difficult to capture every potential outcome. Dispersal, interactions with non-target species, and environmental variability remain significant unknowns. Nonetheless, without models, decision-makers would face an uncertain landscape with limited tools for risk assessment.

Understanding gene drive dynamics requires a range of modelling approaches, each with its strengths and limitations. Some focus on genetic changes within populations, while others incorporate spatial movement, ecological interactions, and stochastic events. Simple models provide clarity and precision, while more complex models capture real-world intricacies.

It is important to note that the organisation of these approaches presented in the following sections reflects my interpretation of the field. While I have drawn from the available literature, this categorisation should be seen as a guide rather than an authoritative classification.

2. Foundational modelling approaches: simplicity for clarity and precision

To understand how gene drives might behave in real-world settings, researchers start with foundational modelling approaches that simplify biological complexity to establish clear expectations. Deterministic models are often the first step. They provide structured predictions by using fixed biological parameters, offering precise insights into how gene drives might spread and persist. These models are excellent for setting baseline expectations but don’t account for random events or variability in natural systems.

Unlike deterministic ones, stochastic approaches consider randomness to capture unpredictable events such as genetic drift or situations where the gene drive might fail to spread. These models add a layer of realism by simulating the uncertainties and variability that are inevitable in biological systems (Combs et al., 2023; Kim et al., 2023). Together, deterministic and stochastic models form a crucial toolkit for researchers, helping them evaluate the feasibility of gene drives and anticipate potential risks and benefits before any environmental release.

2.1. Deterministic, non-spatial models: establishing baseline expectations

Deterministic, non-spatial models are cornerstones for understanding EGD population dynamics. By assuming ideal conditions and focusing solely on genetic processes, these models strip away complexity to offer clear insights. They are particularly valuable for testing hypotheses and gauging gene drive performance in simplified scenarios.

For instance, Zhu et al., (2024) explored how integrating homing gene drives with female-specific lethal alleles could suppress pest populations. Their deterministic model predicted that this combination would significantly enhance population suppression while maintaining self-limitation to avoid unintended spread. This wasn’t just theoretical; experiments with Drosophila melanogaster supported the model predictions, demonstrating high female lethality and a strong inheritance bias.

Similarly, North et al., (2020) used a deterministic, non-spatial model to evaluate a CRISPR-Cas9 gene drive targeting the doublesex (dsx) gene in malaria vectors. The model predicted up to 95% suppression of mosquito populations within four years under ideal conditions. While the study acknowledged factors such as fitness costs and resistance development, it highlighted the importance of deterministic non-spatial models in providing initial proof-of-concept evidence.

These models lay the groundwork for more complex analyses, helping researchers understand the potential of gene drives before considering additional layers of ecological and spatial complexity.

2.2. Expanding with spatial dynamics for real-world applicability

In natural ecosystems, populations live within diverse and changing landscapes where migration, habitat fragmentation, and environmental variability would influence EGD spread. Spatial models address these complexities by adding geographic and movement dynamics, bridging the gap between simplified scenarios and the intricacies of real-world ecosystems (Kim et al., 2023).

For example, North et al. (2020) extended their work to include a spatial component, simulating gene drive spread across West Africa when considering seasonal rainfall patterns and geographic barriers. Their findings revealed significant regional variation in suppression effectiveness, emphasising the need for tailored, location-specific strategies to maximise impact and minimise unintended consequences.

At a continental scale, Beeton (2022) explored the spread of gene drives in malaria vector populations across sub-Saharan Africa. Their simulations incorporated dispersal mechanisms, such as wind-assisted migration, and varying habitat suitability. The model predicted that gene drives could travel from offshore islands to the African mainland within a decade under certain conditions, raising critical questions about risk assessment and the importance of containment measures (Figure 2).

Spatial models are also being used to investigate strategies for containing gene drives. Zhang and Champer (2024) used a reaction-diffusion model to evaluate the ability of TADE suppression drives to remain confined within target populations. By analysing factors such as fitness costs and low-density growth rates, they demonstrated how strategic release patterns could minimise unintended spread while achieving effective suppression of target populations.

What is a reaction-diffusion model?
A reaction-diffusion model describes how biological entities, such as gene drive alleles, spread over space and time through biological interactions (reaction) and movement across landscapes (diffusion).

Understanding the TADE suppression drive
The Toxin-Antidote Dominant Embryo (TADE) suppression drive disrupts an essential gene in the germline while providing a rescue mechanism, ensuring that only gene drive-carrying individuals survive, gradually eliminating the target population.

Despite their utility, spatial models face challenges such as high computational demands and scaling issues. To address this, Champer et al. (2024) proposed an innovative resource-explicit modelling approach. This method replaces direct individual interactions with shared resource nodes, significantly improving computational efficiency without sacrificing ecological accuracy. They tested this approach in a study of rodent populations on New Zealand’s South Island (figure 3), highlighting its potential for modelling larger populations and complex landscapes.

While Champer et al.’s work did not specifically focus on EGD, the resource-explicit approach provides a robust framework for future research, particularly for investigating how gene drives might interact with ecological factors such as resource availability and habitat fragmentation.

Building on the need for scalable and ecologically grounded spatial models, another recent tool takes a different approach to capturing mosquito dynamics and gene drive spread across real-world landscapes.

A recently developed modelling tool, Mozzie, is designed to simulate mosquito lifecycle dynamics along with both local diffusion and wind-assisted long-range dispersal (Wilkins et al., 2025). According to the authors, Mozzie could support the analysis of gene drive spread across diverse ecological contexts, from local environments to continental scales. The tool allows for spatial and temporal variation in parameters such as breeding rates, habitat suitability, and dispersal, which may reflect real-world conditions like seasonal rainfall or daily temperature fluctuations. Mozzie was used by its developers to simulate the spread of a population-modifying gene drive in Anopheles gambiae s.s. and Anopheles coluzzii across sub-Saharan Africa (the same one used by Beeton et al., 2022), incorporating features such as hybridisation, interspecies competition, and landscape constraints. In one example (Figure 4), a gene drive was introduced into one species, with modified mosquitoes released at a central location within a spatial domain divided by a water body. Since mosquitoes could only cross the water via wind advection, seasonal wind patterns influenced the direction and rate of spread. Over time, the gene drive propagated through interbreeding, while resistance emerged in some areas, reducing the modification’s effectiveness. According to the authors, these simulated dynamics may help explore containment strategies and inform large-scale risk assessments.

3. Advanced modelling techniques: addressing uncertainty and complexity

While deterministic and spatial models provide significant insight, they may not capture the inherent randomness and complexity of ecological systems. Advanced modelling techniques, including stochastic and agent-based models, embrace this uncertainty and provide a more nuanced understanding of gene drive behaviour. These approaches enable the simulation of random events, individual behaviours, and complex population dynamics.

3.1. Stochastic approaches: capturing variability and potential risks

Real-world biological systems are rarely as predictable as deterministic models assume. Stochastic models attempt to address this by incorporating randomness, simulating multiple possible outcomes based on factors such as genetic drift, demographic fluctuations, and environmental variability. This approach is particularly valuable when modelling suppression drives, where small population sizes and random events can significantly affect outcomes (Kim et al., 2023).

To facilitate such advanced simulations, Haller and Messer (2019) developed SLiM 3, an advanced software tool designed to expand the capabilities of traditional population genetic models. SLiM 3 incorporates stochastic elements, such as overlapping generations, spatial interactions, and density-dependent regulation, enabling researchers to simulate ecological complexities that deterministic models might overlook. By capturing these complexities, SLiM 3 allows researchers to simulate local extinction and recolonisation events, EGD phenomena that deterministic models might overlook.

An example of leveraging stochastic modelling is the work of Li and Champer (2023), who applied a dual approach to investigate a gene drive system based on Wolbachia cytoplasmic incompatibility genes. Their model combined deterministic equations with stochastic simulations to explore how gene drives persist under different ecological conditions. This approach allowed them to consider both predictable genetic interactions and the uncertainties introduced by real-world factors, including spatial variation and fluctuating population sizes.

A recent study by Camm and Fournier-Level (2025) used a stochastic population genetics model to explore what they called intermediate outcomes in gene drives. An intermediate outcome happens when a gene drive doesn't fully spread through a population but also doesn't disappear completely. Instead, it settles into a middle ground, either becoming temporarily common and then declining or staying stable at moderate levels. The researchers found that achieving these outcomes depends on carefully balancing the effectiveness of the gene drive and the likelihood of resistance evolving. Such outcomes are particularly interesting because they would offer safer and more controlled options, making them potentially valuable for initial gene drive tests or situations where the goal is to limit ecological risks.

3.2. Agent-based models: individual-level simulations for complex interactions

While stochastic models incorporate randomness, agent-based models (ABMs) take this a step further by simulating individual organisms as unique agents with specific behaviours and interactions. This bottom-up approach allows researchers to examine how individual-level decisions and interactions drive population-level patterns.

ABMs are particularly well-suited for studying spatially structured populations, where local behaviours can lead to emergent dynamics that are otherwise difficult to predict (Kim et al., 2023). These models have proven invaluable for understanding the ecological complexities of gene drive systems and evaluating their potential impacts.

Several studies illustrate the utility of ABMs in exploring gene drive behaviour and informing deployment strategies:

Champer et al. (2021) designed an ABM to simulate the spread of suppression gene drives in an island rodent population. Realistic features included overlapping generations, dispersal patterns, and the potential for genetic resistance. Recognising the computational challenges posed by the high number of parameters, they introduced a machine-learning-assisted metamodel. The authors argued that this innovation significantly reduced computational costs while retaining the model predictive accuracy, enabling a comprehensive exploration of ecological and genetic interactions.

Birand et al. (2022) focused on using ABM to investigate the eradication potential of EGD in invasive mice populations on the Antipodes Island. The study revealed the occurrence of "chase dynamics", where local extinction and recolonisation events slowed the eradication process (Figure 5). By simulating spatial dispersal and mating behaviours, the model highlighted how population structure and movement dynamics can impact gene drive efficacy.

Wickramasooriya et al. (2024) applied an ABM to assess the deployment of gene drive mosquitoes on the island of Príncipe, a proposed site for field trials. Their model incorporated environmental factors such as temperature, humidity, and rainfall to simulate mosquito life stages and dispersal patterns. The findings suggested that under carefully planned release conditions, gene drive mosquitoes could achieve a 90% prevalence within 100 days. This study provided essential insights for optimising field trial strategies and understanding environmental influences on gene drive success.

4. Hybrid and decision-support models: bridging genetics with public health goals

As gene drive technologies move closer to field application, there is a growing need for models that integrate genetic, ecological, and public health considerations. Hybrid and decision-support models address this need by combining multiple modelling frameworks, offering a comprehensive perspective of gene drive interventions and their broader implications.

4.1. Hybrid approaches: integrating multiple modelling frameworks

Hybrid models stand out for their ability to merge genetic, ecological, and epidemiological components, enabling a more holistic understanding of how gene drives might influence disease transmission and ecosystems. These approaches are particularly useful for evaluating the feasibility, risks, and benefits of gene drive deployments.

Mondal et al., (2024) exemplified this with their development of MGDrivE 3, an advanced software tool that simulates the epidemiological impacts of gene drive mosquito systems. Unlike traditional models that focus solely on vector population dynamics, MGDrivE 3 integrates entomological and epidemiological factors. The decoupled vector-human approach of the model is particularly innovative, as it simulates mosquito and human populations separately but connects them through shared transmission parameters (Figure 6). This modular framework provides the flexibility needed to adapt to different ecological and epidemiological settings, making it a valuable tool for evaluating gene drive interventions in diverse contexts.

Similarly, Hancock et al., (2024) adopted a hybrid approach to assess EGD interventions across sixteen malaria-endemic regions in West Africa (Figure 7). Their model incorporated ecological, spatial, and epidemiological data to estimate the impacts of gene drives on malaria prevalence and clinical cases. By combining gene drives with existing interventions such as insecticide-treated nets and malaria vaccination, the model showed a potential for substantial public health benefits. Notably, the study highlighted the importance of tailoring strategies to local conditions, which emphasises that deployment success depends on accounting for region-specific ecological and epidemiological variables.

Both studies showcase the power of hybrid models in providing actionable insights.

4.2. Decision-support models for policy and regulation

Decision-support modelling will be essential to potentially turning gene drive research into real-world action. These tools take predictions from different modelling approaches (ecological, epidemiological, and even economic) and package them in a way that policymakers can use to guide the development of regulatory frameworks and deployment strategies.

A standout example of this approach comes from Metchanun et al. (2022). Their study was the first to explore both the epidemiological impact and cost-effectiveness of gene drive mosquitoes for malaria elimination. Using an agent-based model, they simulated scenarios in the Democratic Republic of the Congo, combining gene drive releases with established interventions including insecticide-treated nets and artemisinin-based combination therapy.

One of the key findings was that a single, well-planned release of highly efficient gene drive mosquitoes could eliminate malaria within five years. On top of that, the study also provided economic insights, estimating that the intervention could cost as little as 0.72 to 7.17 International Dollars per person per year, potentially making it the most cost-effective malaria control strategy in resource-limited settings.

By incorporating cost considerations alongside epidemiological projections, this study shows the power of decision-support modelling approaches to guide complex choices.

5. Navigating challenges and future directions in gene drive modelling

Despite the progress made in gene drive modelling, several challenges remain. One of the most pressing is the lack of empirical field data for calibrating and validating these models (Frieß et al., 2023). Unlike pesticide models, for example, which benefit from extensive field trials, EGD models must rely heavily on theoretical assumptions and laboratory experiments. The reliance on non-real-world data introduces unavoidable uncertainty into predictions, complicating both regulatory decision-making and public acceptance.

One promising way to narrow this gap is to connect mosquito genomics more directly with simulation modelling. Rašić and Marshall (2025) recently reviewed how population genomic data shed light on mosquito population structure, levels of gene flow, and standing genetic variation at potential target sites, as well as the sources of resurgence after control efforts. They argue that incorporating these data into models can sharpen parameter estimates and support predictions that are specific to particular regions and vector populations. This approach does not remove uncertainty, but it can anchor models in measurable features of mosquito biology and landscape connectivity, which is valuable for both intervention design and risk assessment.

Another significant challenge lies in balancing model complexity and usability. Adding detailed ecological and evolutionary factors enhances realism, but it also increases computational demands and can make models harder to interpret. Conversely, overly simplified models may fail to capture critical variables, risking inaccurate predictions (Golnar et al., 2021). Striking the right balance is essential to ensure models are both practical and informative.

As the gene drive field moves closer to potential real-world deployment, modelling efforts must evolve. The focus needs to shift from proof-of-concept studies to practical risk assessments that account for diverse environmental conditions and potential unintended effects, such as resistance evolution or unintended gene flow to non-target populations (Combs et al., 2023). This evolution demands closer collaboration between scientists, ecologists, and regulators to ensure models address real-world complexities.

Emerging technologies such as machine learning and network science offer promising tools for overcoming some of these hurdles. These approaches could improve model adaptability, enhance predictive accuracy, and provide more nuanced insights into gene drive spread and containment strategies (Kim et al., 2023). Ensuring consistency and compatibility across different modelling frameworks will also be critical to building regulatory confidence and trust among stakeholders.

However, a less discussed but equally significant challenge lies in the ability of regulatory agencies to understand and apply these increasingly complex models. With gene drive models integrating genetics, ecology, epidemiology, and socioeconomics, how can regulators develop the expertise needed to evaluate them rigorously? And what measures can be taken to ensure that modelling results are accessible and actionable for policy decisions? Addressing this gap will require capacity-building initiatives, clear protocols, and collaborative efforts between modellers and regulators.

Ultimately, the potential success of gene drive technology will depend not only on scientific advancements but also on effective communication with policymakers, stakeholders, and the public. Transparent discussions about model assumptions, limitations, and uncertainties will be key to fostering trust and ensuring responsible deployment.

The following infographic visually summarises how this article organises gene drive modelling approaches. Overlaps exist, reflecting the complexity of gene drive research and the need for flexible modelling frameworks.

Subscribe to stay informed as new evidence reshapes the field: https://monitoringgenedrives.com/contact/

6. References

Birand, A., Cassey, P., Ross, J. V., Russell, J. C., Thomas, P., & Prowse, T. A. A. (2022). Gene drives for vertebrate pest control: Realistic spatial modelling of eradication probabilities and times for island mouse populations. https://doi.org/10.1111/mec.163

Beeton, N. J., Wilkins, A., Ickowicz, A., Hayes, K. R., & Hosack, G. R. (2022). Spatial modelling for population replacement of mosquito vectors at continental scale. https://doi.org/10.1371/journal.pcbi.1009526

Camm, B. J., & Fournier-Leve, A. (2025). Controlling the frequency dynamics of homing gene drives for intermediate outcomes. https://doi.org/10.1093/g3journal/jkae300

Champer, S. E., Oakes, N., Sharma, R., Garcia-Diaz, P., Champer, J., & Messer, P. W. (2021). Modeling CRISPR gene drives for suppression of invasive rodents using a supervised machine learning framework. https://doi.org/10.1371/journal.pcbi.1009660

Champer, S. E., Chae, B., Haller, B. C., Champer, J., & Messer, P. W. (2024). Resource-explicit interactions in spatial population models. https://doi.org/10.1111/2041-210X.14432

Combs, M. A., Golnar, A. J., Overcash, J. M., Lloyd, A. L., Hayes, K. R., O’Brochta, D. A., & Pepin, K. M. (2023). Leveraging eco-evolutionary models for gene drive risk assessment. https://doi.org/10.1016/j.tig.2023.04.004

Frieß, J. L., Lalyer, C. R., Giese, B., Simon, S., & Otto, M. (2023). Review of gene drive modelling and implications for risk assessment of gene drive organisms. https://doi.org/10.1016/j.ecolmodel.2023.110285

Golnar, A. J., Ruell, E., Lloyd, A. L., & Pepin, K. M. (2021). Embracing dynamic models for gene drive management. https://doi.org/10.1016/j.tibtech.2020.08.011

Haller, B. C., & Messer, P. W. (2019). SLiM 3: Forward genetic simulations beyond the Wright–Fisher model. https://doi.org/10.1093/molbev/msy228

Hancock, P. A., North, A., Leach, A. W., Winskill, P., Ghani, A. C., Godfray, H. C. J., Burt, A., & Mumford, J. D. (2024). The potential of gene drives in malaria vector species to control malaria in African environments. https://doi.org/10.1038/s41467-024-53065-z

Kim, J., Harris, K. D., Kim, I. K., Shemesh, S., Messer, P. W., & Greenbaum, G. (2023). Incorporating ecology into gene drive modelling. https://doi.org/10.1111/ele.14194

Li, J., & Champer, J. (2023). Harnessing Wolbachia cytoplasmic incompatibility alleles for confined gene drive: A modeling study. https://doi.org/10.1371/journal.pgen.1010591

Metchanun, N., Borgemeister, C., Amzati, G., von Braun, J., Nikolov, M., Selvaraj, P., & Gerardin, J. (2021). Modeling impact and cost-effectiveness of driving-Y gene drives for malaria elimination in the Democratic Republic of the Congo. https://doi.org/10.1111/eva.13331

Mondal, A., Sanchez C., H. M., & Marshall, J. M. (2024). MGDrivE 3: A decoupled vector-human framework for epidemiological simulation of mosquito genetic control tools and their surveillance. https://doi.org/10.1371/journal.pcbi.1012133

North, A. R., Burt, A., & Godfray, H. C. J. (2020). Modelling the suppression of a malaria vector using a CRISPR-Cas9 gene drive to reduce female fertility. https://doi.org/10.1186/s12915-020-00834-z

Rašić, G., & Marshall, J. M. (2025). Integrating mosquito genomics into simulation modeling: Opportunities for better-informed biocontrol. https://doi.org/10.1016/j.cois.2025.101456

Rode, N. O., Estoup, A., Bourguet, D., Courtier-Orgogozo, V., & Débarre, F. (2019). Population management using gene drive: Molecular design, models of spread dynamics and assessment of ecological risks. https://doi.org/10.1007/s10592-019-01165-5

Wickramasooriya, S., Mahmood, I., Calinescu, A., Wooldridge, M., & Lanzaro, G. (2024). Exploring the dynamics of gene drive mosquitoes within wild populations using agent-based simulation. https://doi.org/10.23919/ANNSIM61499.2024.10732671

Wilkins, A., Beeton, N. J., El-Hachem, M., Hayes, K. R., & Hosack, G. R. (2025). Mozzie: A computationally efficient simulator for the spatio-temporal modelling of mosquitoes. https://doi.org/10.21105/joss.07324

Zhang, S., & Champer, J. (2024). Performance characteristics allow for confinement of a CRISPR toxin–antidote gene drive for population suppression in a reaction–diffusion model. https://doi.org/10.1098/rspb.2024.0500

Zhu, J., Chen, J., Liu, Y., Xu, X., & Champer, J. (2024). Population suppression with dominant female lethal alleles is boosted by homing gene drive. https://doi.org/10.1186/s12915-024-02004-x