Dual‐fluorescence imaging and automated trophallaxis detection for studying multi‐nutrient regulation in superorganisms
Cite this dataset
Baltiansky, Lior et al. (2021). Dual‐fluorescence imaging and automated trophallaxis detection for studying multi‐nutrient regulation in superorganisms [Dataset]. Dryad. https://doi.org/10.5061/dryad.j9kd51cc7
Abstract
In the related manuscript, we present a dual-fluorescence imaging setup designed to track two food sources, each labeled with a different fluorophore, as they are disseminated throughout a freely behaving colony of individually tagged ants. Additionally, our image-based deep learning algorithm for automatic detection of ant trophallaxis events efficiently yields a detailed record of all food-transfer interactions.
Using a series of calibration experiments, we demonstrate the reliability of our measurements. We then exemplify the capabilities of our new method by tracking food dissemination in a colony of Camponotus sanctus ants supplied with two nutritionally-distinct food sources.
This dataset contains data and Matlab code related to:
1. Calibration and validation of the dual-fluorescence imaging technique
2. Training and employing a deep neural network for detecting trophallaxis
3. Sample data from a multinutrient feeding experiment
Methods
This data was collected using the experimental setup described in the manuscript.
For further details, see README files within and related manuscript.
Usage notes
Dual fluorescence imaging:
Data presented here was used to generate the figures in the related manuscript.
Trophallaxis detection:
We here provide our trained trophallaxis detection network and Matlab code for using it to detect trophallaxis in videos of Camponotus sanctus ants tagged with Bugtag (Robiotec) tags.
One may modify the code to apply to videos of ants tracked with a different method (e.g. other tags, tracking of untagged ants), or to videos of untracked ants if the identity of ants is not important to one's needs.
As the supplied network has been trained only on images of C. sanctus ants, it may be less optimal for detecting trophallaxis in other species. One may use the supplied network as a starting point for training detection of trophallaxis of other species. We supply code for transfer learning which was used for training our network. This code can be modified for retraining our network on additional training data.
We also provide the library of annotated images used to train the trophallaxis detection network. One may use these images for training networks and classifiers other than the one we provide.
Funding
Israel Science Foundation, Award: 1727/20
Estate of Seymour Rosenwasser
Dorith Amos Trust
Estate of Robert A. Moss
Estate of Robert Einzig
Clore Center for Biological Physics
Henry J. Leir Professorial Chair