Predicting moqsuito flight behavior using Bayesian dynamical systems learning

This is the official repository for the paper Predicting mosquito flight behavior using Bayesian dynamical systems learning. We combine 3D infrared tracking of Aedes aegypti with Bayesian dynamical systems inference to learn quantitative biophysical models of mosquito host-seeking behavior.

The data is hosted on zenodo. To download the dataset, we have provided a bash script scripts/download_data.sh. You can run this script with

bash scripts/download_data.sh [DESTINATION_DIR] 

DESTINATION_DIR is the directory to save the downloaded files. The default is data, and we recommend leaving this as some of the provided code uses this directory.

To set up the Julia package with the necessary dependencies, run

julia scripts/setup.jl

NOTE: You must have Julia downloaded and installed. You can install the latest version here. We recommend having at least version 1.11.

To filter and visualize our datasets, check out the examples/00_data_viz.ipynb notebook. Make sure to download the data in the data directory before running this notebook.

To see a walk through of how to perform inference on a simulated system of mosquitos without any sensory cues, see examples/01_Sim_NoCues.ipynb. In this notebook, we recreate Figure S3 in our paper.

To see a walk through of how to perform inference on a simulated system of mosquitos that are repelled from a center point, see examples/02_Sim_Repellent.ipynb. In this notebook, we recreate Figure S4 in our paper.

To see a walk through of how to perform inference on a mosquito dataset with visual and CO₂ cues, see examples/03_Real_Inference.ipynb. In this notebook, we recreate Figure 4 B and C in our paper.

To see a walk through of how to simulate the inferred mosquito models and generate synthetic trajectories, see examples/04_Real_LearnedModels.ipynb. In this notebook, we create plots similar to Figure 5 C, F, I, and L of our paper.

We developed an interactive web application for users to explore our models. The application allows users to specify which sensory cues are present, originating at the center of the image in the screen. Mosquitos move in real time and the user can move the image around to change the position of the cues. Users can also replace the default image of Tim the Beaver for fun!

We would like to thank Dr. Ellen Dotson and Laura Leite of the CDC, and Catherine Steele from the CDC Foundation for discussions and their work rearing and raising the mosquitoes. We would like to thank Dr. James Sutcliffe (Emeritus, Trent University), for discussions on mosquito behavior, operating the photonic sentry and explaining his experimental set up. We also appreciate Alison Wang’s assistance in running the experiments with Anopheles mosquitoes.

Please cite our paper if you find it useful for your work.

@article{zuo2026predicting
	doi:10.1126/sciadv.adz7063,
	author = {Christopher Zuo  and Chenyi Fei  and Alexander E. Cohen  and Soohwan Kim  and Ring T. Cardé  and Jörn Dunkel  and David L. Hu },
	title = {Predicting mosquito flight behavior using Bayesian dynamical systems learning},
	journal = {Science Advances},
	volume = {12},
	number = {12},
	pages = {eadz7063},
	year = {2026},
	doi = {10.1126/sciadv.adz7063},
	URL = {https://www.science.org/doi/abs/10.1126/sciadv.adz7063},
	eprint = {https://www.science.org/doi/pdf/10.1126/sciadv.adz7063},
}