ausblenden:
Schlagwörter:
-
Zusammenfassung:
Neural networks working near criticality (a transition point between ordered and chaotic states) have been demonstrated to possess optimized information processing capabilities [1,2]. One way to assess criticality in neural networks is to study spatiotemporal patterns of activity propagations known as neuronal avalanches. At criticality, the avalanche sizes follow a power-law distribution up to a cutoff approximately at the system size [3]. The conventional method of detecting avalanches assumes they are separated in time. However, in large-scale recordings where many avalanches happen in parallel, this method cannot be applied. To asses the state of the system in such a case, one can use a cluster-based avalanche detection that can spatially segregate parallel avalanches assuming they propagate locally [4]. This method revealed that the zebrafish larvae brain operates close to criticality in the absence of stimulus and is transiently displaced to a slightly more ordered regime upon presenting moving grating [5]. However, how more complex stimuli change the brain state of the zebrafish larvae is unknown.
Here, we apply cluster-based avalanche detection to larval zebrafish pretectum calcium imaging data, which was obtained during the presentation of contiguous motion noise visual stimuli [6]. We assume that each neuron is locally connected to other neurons within a detection-radius. We find that choice of the detection-radius largely influences the avalanche-size distributions (Fig A), and only some radii give rise to power-law distributions extending to system size (Fig B).
To understand the effect of neighborhood definition on observed avalanches, we simulate a branching network model [7] with local connections and generate parallel avalanches by injecting random stimuli at multiple sites. At criticality, the avalanche-size distributions are independent of the detection-radius and exhibit power-law distributions extending beyond the system size (Fig C). On the other hand, at the subcritical regime, avalanche-size distributions strongly depend on the detection-radius, though some power-laws can still be observed (Fig D). These results suggest that measuring the divergence of distributions under different choices of neighborhood provides additional insights in assessing the state of neural networks when the connectivity structure is unknown. They also propose that under complex visual stimuli the zebrafish larvae brain exhibits subcritical dynamics.
