Bimodal Von Mises Mixture Model for Locust Directional Preferences

This repository contains a bimodal von Mises mixture model developed to analyze the directional preferences of individual locusts in response to virtual swarms of conspecifics in virtual reality (VR). The model supports the findings presented in Sayin et al., 2025, which challenge classical models of collective motion and propose a minimal cognitive framework for locust movement.

Overview

The bimodal von Mises mixture model is designed to:

• Quantify directional preferences of individual locusts exposed to VR-generated swarms.
• Fit a probabilistic model to the observed data, capturing both individual and population-level responses.
• Test hypotheses on decision-making mechanisms, particularly in relation to the vectorial representation of neighbors versus optomotor responses.

Features

• Bayesian Inference: Utilizes PyMC for probabilistic modeling of locust movement.
• Circular Statistics: Applies von Mises distributions to analyze periodic directional data.
• Preprocessing Pipeline: Rediscretizes locust trajectories to minimize auto-correlation.
• Visualization Tools: Generates trajectory plots and statistical summaries using matplotlib and seaborn.

Context & Relevance

This model is part of a broader study on the behavioral mechanisms of collective motion in desert locusts, as described in Sayin et al. (2025). The study demonstrates that locusts do not rely on explicit alignment with their neighbors but rather use an internal decision-making framework based on neural representations of bearings and social interactions.

Key findings that this model contributes to include:

• Directional bias analysis: Testing whether locusts prefer certain bearings in response to stimuli.
• Assessment of individual variability: Estimating the extent to which different locusts exhibit unique movement tendencies.
• Evaluation of cognitive models: Providing empirical evidence against classical self-propelled particle models and supporting a vectorial decision-making approach.

Usage

The Jupyter Notebook (locust_mixture_model.ipynb) contains:

• Data preprocessing: Rediscretization of locust trajectories to reduce autocorrelation.
• Probabilistic modeling: Implementation of Bayesian inference using PyMC.
• Visualization: Plotting movement trajectories and fitting mixture models.
• Statistical inference: Estimating posterior distributions of directional preferences.

Citation

If you use this model, please cite:

@article{sayin2025behavioral,
  title={The behavioral mechanisms governing collective motion in swarming locusts},
  author={Sayin, Sercan and Couzin-Fuchs, Einat and Petelski, Inga and G{"u}nzel, Yannick and Salahshour, Mohammad and Lee, Chi-Yu and Graving, Jacob M and Li, Liang and Deussen, Oliver and Sword, Gregory A and others},
  journal={Science},
  volume={387},
  number={6737},
  pages={995--1000},
  year={2025},
  publisher={American Association for the Advancement of Science}
}

Data Availability

The data used in this study are not yet available but will be released at the following Zenodo repository: https://doi.org/10.5281/zenodo.14353283

License

This repository is licensed under the Apache 2.0 License.