Stochastic modelling of mosquito behaviour in response to environmental conditions

BackgroundUnderstanding mosquito foraging behaviour is of fundamental biological interest and practical importance for preventing infectious diseases such as malaria and dengue. Mosquitoes sense CO2, skin odours, heat and moisture with high sensitivity, but how mosquitoes use this information to navigate towards a human in turbulent air flow is yet unclear. Tracking experiments in wind tunnels are an advanced model to study mosquito host seeking flight but have important limitations and the results from current regression-type analyses are hard to generalise to field settings. In addition, such analyses typically rely on summary statistics of whole trajectories and thus don't account for autocorrelation and behavioural changes within trajectories. Therefore, new analytic frameworks and predictive model for individual mosquito flight trajectories are needed to advance understanding of mosquito host seeking behaviour. AimsI aim to improve the understanding of mosquito foraging flight in response to the environmental conditions with respect to airflow, CO2, skin odour and heat. In particular, I will (A) Provide a framework to analyse mosquito flight on the level of individual trajectories and simulated data on airflow, CO2 and odour along each trajectory; (B) Model mechanistically how mosquitoes use sensorial information on air velocity, CO2 odour and heat to navigate during foraging flight; (C) Asses to what extent the experimentally observed mosquito foraging flight trajectories are driven by cue-following and identify flight patterns which relate to other behavioural modes.MethodsI use computational estimators of intrinsic dimension (ID) to analyse behavioural modes within single trajectories. I use Ornstein-Uhlenbeck (OU) stochastic processes and modifications thereof, which depend on environmental covariates as predictive models of mosquito movement at the trajectory level, and fit them to data by computational Bayesian methods. Depending on the success in addressing the objectives with the OU approach and depending on timing, I may also use stochastic simulators fitted to data by approximate Bayesian computation and a machine learning simulator trained with reinforcement learning. The lab experiments that generated the tracking data and the computational fluid dynamics simulations that generated the data on environmental conditions were performed by collaborators.ResultsI expect to deliver an analytic framework based on intrinsic dimension to cluster behavioural modes within single trajectories. Thus, I will be able to quantify for each trajectory the alignment of behavioural switches with changes in the local environment. I will likely find that mosquitoes enter and exit cue-following behaviour and I will be able to indicate under what environmental conditions and in what areas of the wind tunnel this is likely to happen. I have already partly succeeded in fitting of a stochastic movement model to tracking data to quantify the attractiveness of the cue source while taking into account the full time-series data. It is difficult at present to forecast how well the proposed predictive movement models fit to the real data. However, I will likely be able to test whether the hypothesis that a fixed sequence of behavioural modes depending on certain cue combinations guides host seeking flight is supported by the experimental data. ImpactA predictive model of mosquito flight trajectories depending on high-resolution environmental conditions is a flexible tool to optimise design of mosquito control interventions such as odour-baited traps. Thus, my work has potential for an important public health impact. In addition, my approach in modelling animal movement in response to high-resolution environmental conditions may be extended to other species.