Review of gene drive modelling and implications for risk assessment of gene drive organisms

•Gene drive models to date focus more on drive efficacy than ecological realism.•The majority of models do not incorporate a specific target organism's life history.•Multiple potentially decisive ecological factors are rarely included.•To support an environmental risk assessment more ecological realism would be needed. Synthetic gene drive (GD) systems constitute a form of novel invasive environmental biotechnology with far-reaching consequences beyond those of other known genetically modified organisms (GMOs). During the last 10 years, the development of GD systems has been closely linked to mathematical modelling which can provide feedback on how to achieve gene drive spread but also may be used to predict the ecological consequences of a gene drive release. GMOs, thus also GD systems, need to pass an environmental risk assessment (ERA) prior to a release into the environment. Models in this respect may play an important role because a release of GD organisms, even at a small scale, may not be reversible. In our review, we analyse the scope and structure of existing models to examine how they may assist the ERA. Our analysis reveals that a majority of models so far are deterministic, non-spatial and not tailored for a specific target organism. Models often use simplified assumptions on the biology of the species and seem to be made to test the effectiveness of the drive. Few models go beyond this and verify whether model predictions may be realistic under field conditions. We identified four advanced models that we judged to be the most ecologically realistic and compared the implemented parameters with ERA requirements by the European Food Safety Authority (EFSA) and World Health Organization (WHO) for genetically modified insects and mosquitoes. Although a number of abiotic and biotic factors are already considered in these models, mating-related factors and traits relevant to the interactions between the GMO and target organisms and with other species are largely excluded. Overall, our results show that biological and ecological realism are still poorly realized in current models and that most models aim to predict efficacy rather than ecological effects. Given the complexity of natural ecosystems, it may not be possible to compile a single model to cover all complexities. Thus, models should be further developed with the purpose to assist specific questions related to the risk assessment of GDs. Moreover, uncertainty will be a key issue for any model