P273 A three-state model for the temporal statistics of mouse behavior

Friedrich Schuessler*1, Paul Mieske2, Anne Jaap3, Henning Sprekeler1

1 Department of Computer Science, Technical University Berlin, German
2German Center for the Protection of Laboratory Animals (Bf3R), German Federal Institute for Risk Assessment, Berlin, Germany
3Department of Veterinary Medicine, Free University of Berlin, Germany

*Email: f.schuessler@tu-berlin.de


Introduction
Neuroscience is undergoing a transition to ever larger and more complex recordings and an accompanying surge of computational models. A quantitative or computational description of behavior, in contrast, is still in dire need [1]. One important aspect of behavior is the temporal structure, which contains rhythmic components (circadian), exponential components with specific time scales (duration of feeding), and components with scale-free temporal dynamics (active motion). Understanding better how these aspects arise and interact, both in the individual and within a group of animals, is an important stepping stone towards computational models of behavior.
Methods
Here we analyze the temporal statistics of behavior of mice housed in different environments and group sizes. The main analyses are based on RFID detections of antennae placed throughout the housing modules. We make particular use of the statistics of inter-detection intervals (IDIs).
Results
We find that behavior spanning seconds to hours can be separated into three distinct temporal ranges: short (0-2min), intermediate (2-20min), long (>20min). IDIs for intermediate and long ranges follow two distinct exponential distributions. Short IDIs are more consistent with a power law or mix of multiple time scales. Blocks of successive short IDIs also follow an exponential distribution. We introduce a simple Markov model that reproduces the temporal statistics.Using additional video recordings, we link the temporal regimes to behavior: Short IDIs to explorative or interactive behaviors, intermediate IDIs to feeding and grooming, and long IDIs to sleeping.
Discussion
Our results show a surprisingly simple structure: Behavior on a fast time scale is interrupted by Internal demands on slower time scales: bouts of fast activity are cut off by the need to feed, and longer sequences of activity and feeding are interrupted by the need to sleep. The short-time aspects of behavior match with observations of scale-free statistics in previous studies [2, 3], but also show interesting deviations potentially due to the interactions in the group. Taken together, our results open up the possibility to understand behavior through the lens of simple models, and raise questions about the neural mechanisms underlying the observed structure.

Acknowledgements
We are grateful for funding by the German Research Foundation (DFG) through the Excellence Strategy program (EXC-2002/1 - Project number 390523135).
References
[1] Datta, S. R., Anderson, D. J., Branson, K., Perona, P., & Leifer, A. (2019). Computational neuroethology: a call to action.Neuron,104(1), 11-24.
[2] Nakamura, T., Takumi, T., Takano, A., Aoyagi, N., Yoshiuchi, K., Struzik, Z. R., & Yamamoto, Y. (2008). Of mice and men—universality and breakdown of behavioral organization.PLoS one,3(4), e2050.
[3] Bialek, W., & Shaevitz, J. W. (2024). Long timescales, individual differences, and scale invariance in animal behavior.Physical review letters,132(4), 048401.