Algorithms of information transfer between zebrafish individuals: from perceptual decisions to collective behavior

Informazioni relative al progetto

INTERZEB

ID dell’accordo di sovvenzione: 101153670

DOI

10.3030/101153670

Progetto chiuso

Data della firma CE 15 Aprile 2024

Data di avvio 1 Giugno 2024

Data di completamento 31 Maggio 2026

Finanziato da

• Marie Skłodowska-Curie Actions (MSCA)

Costo totale

Nessun dato

Contributo UE

€ 173 847,36

Coordinato da

UNIVERSITAT KONSTANZ
Germany

Periodic Reporting for period 1 - INTERZEB (Algorithms of information transfer between zebrafish individuals: from perceptual decisions to collective behavior)

Periodo di rendicontazione: 2024-06-01 al 2026-05-31

Sintesi del contesto e degli obiettivi generali del progetto

Rapid information exchange between individuals is key to coordinated behavior. In a traffic lane, a well-timed brake can signal the presence of an obstacle to the drivers behind, avoiding a vehicle pileup. In a crowd of people, individuals can infer the position of an obscured exit from the movement of their neighbors. How do individual behavioral patterns enable efficient information transfer in a group? Animal biodiversity displays a vast amount of remarkable collective phenomena that can provide us insight into this question, such as the behavior of fish shoals. In zebrafish, the trajectories of individuals at the front of the shoal ("leaders") may transmit relevant cues to other members ("followers"). The INTERZEB project uses zebrafish as a model system to experimentally establish the behavioral algorithms that enable individuals to transmit information within a group. Inferring individual behavioral rules within a group has proved extremely difficult in the past: in fact, when looking at a group from outside, one cannot univocally identify the source of information which has triggered a specific action in an individual. In this project, leveraging the visual behavior of juvenile zebrafish, we use virtual reality techniques to experimentally control the flow of information between freely-swimming fish interacting through virtual “avatars”. Knowing “who sees whom” allows us to track the exact connections between fish and allows us to measure information flow across steps. This approach further enables us to estimate which navigation algorithms individual fish use to read-out information and propagate it to others. The achievement of the project’s objectives will establish an unprecedented link between individual sensorimotor control and collective animal behavior, and will offer potential inspiration for biomimetic technologies.

Lavoro eseguito dall’inizio del progetto fino alla fine del periodo coperto dalla relazione e principali risultati finora ottenuti

All experiments were performed in juvenile zebrafish (Danio rerio) at 24-30 days post fertilization (dpf). At this age, zebrafish exhibit a strong attraction towards conspecifics which is visually mediated, offering an outstanding opportunity to make use of visual social stimuli. To experimentally control information flow between multiple individuals, we employed 3D immersive virtual reality (VR) for freely-swimming zebrafish. Each fish was hosted inside a VR apparatus and presented with virtual "avatars" of selected conspecifics. This setup allowed us to propagate information along experimentally-defined social transmission chains involving several fish. Transmission chains were implemented in multiple configurations, including one-to-one and two-to-one transmission steps. In a one-to-one transmission step, a single virtual avatar is shown to a real fish; in a two-to-one step, two virtual avatars are shown simultaneously to a real fish. These transmission steps can be freely combined to generate more complex transmission chains involving multiple steps. For example, an indirect connection can be established between two fish A and C, by showing the avatar of fish A to a fish B in one setup, and the avatar of fish B to fish C in a separate setup. Two transmission chains can be further merged with a two-to-one step, presenting two virtual avatars stemming from different chains to a single fish, making it possible for the fish to combine information from both sources.
We quantified the amount of information lost at each step of the chains using approaches based on both single features (e.g. turn angle) as well as multiple features combined with machine learning approaches. Complementing the empirical observations, we also simulated transmission chains with artificial agents that behaved according to different types of navigation algorithms including control-theoretical models and ring attractor networks. Finally, some of the algorithms were embodied into artificial virtual agents that replaced the real fish at selected steps within the transmission chains.
Overall, we generated and documented high-quality datasets containing tens of thousands of leader-follower events in groups of zebrafish with fully controlled connectivity patterns; we performed quantitative analyses of pursuit behavior and information transfer within the groups; we implemented and validated algorithms to simulate navigation behavior in the data.

Progressi oltre lo stato dell’arte e potenziale impatto previsto (incluso l’impatto socioeconomico e le implicazioni sociali più ampie del progetto fino ad ora)

Information decays non-linearly along transmission chains. At different steps of the chain, different types of navigation controllers are required to capture the fishes’ following behavior effectively. In general, when comparing the transmission chains of fish to those of artificial agents, ring attractor networks outperform simpler control-theoretical models in explaining the following behavior of real fish. Modeling fish behavior with ring attractor networks provides the benefit of offering a biologically-inspired model architecture that is based on, and may directly inform, research on the neuronal underpinnings of navigation behavior.
We further observed that real fish can combine multiple independent sources of information (i.e. when following multiple leaders), leading to reduced information loss along the chain. This behavior can be simulated by a ring attractor network following multiple targets.
Overall, the study 1) advances our understanding of the computations that the brain must implement in order to enable group information processing and navigation, and 2) demonstrates a new experimental framework for explaining individual sensorimotor algorithms in terms of their biological function. After further analysis and validation, the listed results will be published in peer-reviewed venues and communicated to the public through the appropriate channels. Based on the current outcomes, further research can be directed to testing the relevant mechanisms in larger-scale collectives, investigating their neuronal correlates, and implementing them in embodied artificial agents and/or vehicles.