Introduction

The central goal of systems neuroscience is to explain the neural underpinnings of behavior. To investigate the link between sensory input, brain activity and animal behavior, relevant behavioral variables have to be recorded and quantified. Therefore, the same experimental paradigm has to be replicated in different experimental setups in order to combine it with different recording or stimulation techniques, and it needs to be reproducible across different laboratories. However, the setups generally rely on heterogeneous hardware and custom-made software tailored to the specific requirements of one experimental apparatus. Often, the code used is based on expensive software packages (such as LabView or Matlab), with open-source options for hardware control generally limited to one particular type or brand of devices. As a consequence, the same experimental protocol has to be implemented many times, thus wasting time and increasing potential sources of error. This makes sharing the code for replicating a scientific finding under the same experimental conditions very difficult.

To address these problems, we developed Stytra, a package that encompasses all the requirements of hardware control, stimulation and behavioral tracking that we encounter in our everyday experimental work. Our system, completely written in Python, provides a framework to assemble an experiment combining different input and output hardware and algorithms for online behavioral tracking and closed-loop stimulation. It is highly modular and can be extended to support new hardware devices or tracking algorithms. It facilitates reuse of different components of the package, encourages building upon existing work and enforces consistent data management. The definition of experimental protocols in high-level Python scripts makes it very suitable for version control and code sharing across laboratories, facilitating reproducibility and collaboration between scientists. Finally, it runs on all common desktop operating systems (Windows, MacOS and Linux), therefore incurring no additional costs on the software side. Similar approaches have already been made available for real-time tracking of zebrafish larvae [Hae19], [DvDLA18]. Still, to our knowledge, none of these solutions implement tracking functions for both head-restrained and freely-swimming larvae, they do not allow the use of custom tracking algorithms, and they do not provide a generic framework to design open- and closed-loop stimulation paradigms.

Stytra was developed primarily in the context of a laboratory working with larval zebrafish, and it fulfills the common requirements of behavioral paradigms used with this animal [PE09]: video tracking, visual stimulation and triggering of external devices. The tracking functions (for freely swimming and head-restrained fish) include both efficient re-implementations of published algorithms and newly-developed methods. Nevertheless, custom methods can easily be added. Common visual stimuli and methods for combining them and presenting them in different ways are provided. Our experimental setups are open-source as well :cite`chagas2018openhw`: hardware designs provided along with the documentation describe the apparatus required for performing common behavioral experiments in zebrafish in detail. The library provides many elements useful for designing behavioral experiments in Python, potentially offering a unified platform to build and share experiments in zebrafish neuroscience and behavioral research. We welcome and will support community contributions to expand the capabilities of the package to other paradigms and animals, although our development efforts will remain focused on zebrafish applications.