Session N36: Animal Behavior and Social Interactions II

Global and local approaches to detect subtle behavioural changes in large-scale behavioural screens

The central nervous system produces diverse behaviours, like muscular responses, observable via video recordings. Current advancements in genetics, large-scale behaviour tracking, and machine learning facilitate the understanding of how behaviour and neural activities correlate. In organisms like the Drosophila larva, it's now feasible to map this at large scales, covering millions of animals and individual neurons. This enables the pinpointing of neural circuits linked to specific behaviours. High-throughput behavioural screens are invaluable as they relate the activation or deactivation of neurons to behaviour sequences in millions of organisms, showcasing the vast range of neural responses to a single stimulus. However, extracting nuanced behaviours from these screens and interpreting them at a broader scale remain challenges.

We propose multiple robust methods to mine detailed information from behavioural screens:

Constructing motion spaces of ant mandibles from microCT scans using vector embeddings.

Recent advances in microCT scanning technology allow for a large amount of high-fidelity geometric data to be collected across the tree of life. However, these high fidelity 3D scanning methods have rarely been utilized to shed light on the motion or kinematic state space of animals. This imaging methodology is particularly promising for organisms that operate at a scale difficult to record with standard motion capture solutions. Taking ants as a motivating example, we set out to characterize the range of three-dimensional kinematics of ant mandibles across several ant families and among several Hymenopteran clades for a broader phylogenetic context. We make use of microCT scan data of both closed and open states of the mandible of preserved specimens to generate three-dimensional models of both the head and the mandible for a phylogenetically and morphologically broad sample. With these models, we are able to perform three-dimensional interpolation between these open and closed states. Interpolation is done by considering the mandibles as vector embeddings in three-dimensional space. These kinematic predictions are made biologically relevant through collision checking of the mandible geometry against the head geometry. Using the vector embedding representation, we are further able to quantify range-of-motion for use in a comparative framework. Our methodology opens up a new avenue of research based on high-fidelity scanning methods and gives new insight into the kinematics of small-scale organisms.   

Long timescale structure and scale invariance in animal behavior

New continuous, high-dimensional behavioral data from freely moving fruit flies spanning out to time scales of weeks allows for the investigation of circadian variations in stereotyped behaviors and age dependent effects, as well as searching for previously unidentified long time scale dynamics. We quantified behavior using a deep learning framework for pose-estimation (SLEAP)² and spectral clustering (MotionMapper)³, and observe that a fly’s trajectory in state space is strongly non-Markovian⁴ with correlations across many timescales. We propose an analysis scheme that deals with the challenges of non-independence of samples and individual differences to extract long-ranged correlations in behavior trajectories. Scale invariance in behavior with a precise match in scaling using different correlation functions emerges from our analysis and this scaling is seen across multiple decades, ranging from seconds to an hour⁵. We implement this analysis scheme on data spanning weeks to test scaling, determine the limits of scale invariance and estimate the individuality in flies while accounting for aging and circadian patterns.

1. McKenzie-Smith, G. C., et al. arXiv:2309.04044 (2023)

2. Pereira, T.D., et al. Nat Methods, 19.4 (2022): 486-495.

3. Berman, G.J., et al. J R Soc Interface, 11.99 (2014): 20140672.

4. Alba, V., et al. arXiv:2012.15681 (2020)

5. Bialek, W. and Shaevitz, J. W. arXiv:2304.09608 (2023)   

Controlling Noisy Herds

Individual sheep instinctively flee from shepherd dogs, yet in larger groups, they adopt a ‘selfish herd strategy,’ clustering together to minimize risk. This well-known collective behavior falters in small groups (N<5), where sheep unpredictably oscillate between fleeing and clustering. In this talk, we will discuss the physics of controlling such small N, noisy collectives, using the interaction between sheep and shepherd dogs as a case study. We focus on two key tasks: ‘the drive,’ which involves herding the group of sheep together, and ‘the shed,’ the act of splitting the group into two subgroups. Through mathematical modeling and field observations, we show that the key to controlling these erratic small groups lies in exploiting their inherent stochasticity. Our mechanistic framework not only offers a deeper understanding of animal herd dynamics but also provides valuable insights into controlling dynamic and heterogeneous systems across various disciplines, from swarm robots to public policy.    

Environmental Adaptations in Fire Ant Mound Construction

Fire ants are highly invasive and robust social insects that thrive in diverse environments well beyond their native range. Colonies build elaborate subsurface nests by removing soil pellets that are deposited above ground. A surface mound with complex internal structures is eventually formed, although its construction and function are not well understood. We hypothesize that fire ants' success is linked to above- and below-ground architectural adaptations across environmental conditions. To begin to investigate how soil moisture and temperature influence nest structures, we focus on incipient mound formation in the lab. Using a custom-built surface scanner, we acquire high-resolution, time-resolved 3D height maps during the first 24 hours of construction in naturalistic soil. We find that surface features are qualitatively different across soil moisture contents: in wetter soil, ants build many tall and narrow vertical towers sometimes connected by bridges, whereas in drier soil, ants build a few broad, shallow heaps. Surface features are similar across temperatures, though construction rates tend to be higher in warmer conditions. Our results demonstrate that environmental factors influence mound construction, and future work will investigate the origins of this environmentally driven structural variation as well as the interplay between substrate physics and collective behavior in mound formation.   

Coarse-graining non-stationary dynamics: linking early to late phases of zebrafish contests

Long-time behavior in complex systems, such as oceans, climate, and biological processes, is often coupled with non-stationary effects. In animal behavior, slow changes can occur without external stimuli (e.g. through the modulation of internal states such as satiety) and across multiple time scales, challenging their analysis from observed data. Here, we adopt a transfer operator approach to identify coarse-grained modes of non-stationary dynamics in the pair fighting behavior of zebrafish. Using high-resolution body point trajectories, we construct a discrete approximation of the transfer operator through a transition matrix over a waiting time τ. The dominant singular values of the transition matrix are related to the slowest relaxation times of a forward-backward process, and the corresponding left and right singular vectors detect initial and final coarse-grained behavioral states, respectively, "coherent pairs". By varying τ, we can extract coherent pairs across timescales, and we show that fight dynamics exhibit relaxation processes of at least the same duration as the fight itself. We find that the dominant coherent pairs are chasing maneuvers that evolve from those instigated by the eventual loser to those instigated by the winner.   

A physical theory for social behavior

Intercommunity social interactions in organisms from bacteria to humans can drive both mobility and proliferation. Due to the complexity of the individual constituents, construction of continuum models of behavior is difficult. Nevertheless, the existence of collective phenomena such as propagating waves and phase separation suggest that such hydrodynamic theories may be applicable. Here, we combine data-driven techniques with analytical tools from statistical physics to illustrate how to construct continuum models of social behavior. First focusing on socially driven motility, we consider human residential dynamics. Using US Census data, we find that human populations evolve over long length- and time-scales and, with the aid of machine learning, we confirm that the dynamics are local. By modeling humans as utility maximizers, we construct a minimal hydrodynamic theory for interacting human populations and study the effects of "nudging" group preferences on segregation. Finally, we extend our theory to ecological interactions, connecting the effects that utility (or fitness) has on motility as well as birth and death of individuals.   

Dynamics, Statistics, and Task Allocation of Foraging Ants

Ant foraging is one of the most fascinating example of cooperative behavior observed in nature. It is well studied from an entomology viewpoint but there is currently a lack of mathematical synthesis of this phenomenon. We address this by constructing an ant foraging model that incorporates simple behavioral rules within three task groups of the ant colony during foraging (i.e., foragers, transporters, and followers), pheromone trails, and memory effects. The motion of an ant is modeled as a discrete correlated random walk, with characteristic zigzag path that is congruent with experimental data. We simulate the foraging cycle, which consist of ants searching for food, transporting food, and depositing chemical trails to recruit and orient more ants (en masse) to the food source. This allows us to shed insights into the basic mechanism of the cooperative interactions between ants and the dynamical division of labor within an ant colony during foraging to achieve optimal efficiency. We observe a disorder-order phase transition from the start to the end of a foraging process, signaling collective motion at the population level. Finally, we present a set of time delay ODEs that corroborates with numerical simulations.   

Uncovering principles of long timescale sensory evoked navigation in larval zebrafish

Animals navigate their environments by chaining short timescale bouts into long strategies, reflecting the complex interplay between sensory cues and internal states. We study the principles driving these strategies in larval zebrafish by quantifying behavior across timescales. By concatenating the fish’s tail pose across bouts over time, we constructed maximally predictive sequences and investigated the time evolution of behavior through a high-fidelity Markovian model. The eigenspectrum of the inferred transition matrix revealed a hierarchy of timescales with three main modes - the first involving Cruising and Wandering strategies with steady or rapid changes in orientation, the second encoding overall speed, and the third encoding preference for direction (left vs. right).

This approach allowed us to study of how sensory conditions impact navigation along these modes. Aversive stimuli such as expanding spots trigger fast wandering strategies, while chasing dots only reduce speeds. When hunting prey, eye convergence events elicit a switch to slow cruising, whereas inter-hunt behavior is dominated by wandering strategies. Our ensemble model also revealed individual preferences for strategies at the timescale of the experiment, suggesting the influence of long lasting latent states on the behavior overruling the environmental sensory cues. Altogether, our approach illuminates the modulation of behavior by sensory stimuli and individual preferences across a hierarchy of timescales.