Critical Brain Dynamics May Be More Prevalent Than Previously Thought

A new study reveals that human and animal brains may operate in a state of “criticality” much more frequently than previous evidence suggested. The research demonstrates that critical and scale-invariant dynamics—fundamental for the optimization of neural processing—are often hidden by traditional data analysis methods, leading to false negative results. Instead of the brain being far from criticality when data lacks clear power laws, critical activity might simply be camouflaged in low-dimensional subspaces undetectable by conventional analytical approaches.

The Population Average Paradigm The hypothesis that the brain operates near a critical phase transition has been widely supported by the observation of scale-invariant activities, commonly evaluated through neuronal avalanche analysis and long-range temporal correlations (LRTC). The central problem lies in the fact that the overwhelming majority of these empirical studies reduce the activity of thousands of neurons to a single one-dimensional time series based on the population average. When populations of neurons fire in an anticorrelated manner, these dynamics cancel each other out in the global average. Consequently, when these average fluctuations do not fit power laws or when strong oscillations dominate the signal, the scientific community tends to mistakenly reject the criticality hypothesis, assuming the system has drifted from its optimal state.

Unveiling Hidden Criticality with PCA To prove that criticality can be hidden, researchers used a high-dimensional computational model that demonstrated how neural systems separate different dynamical modes into distinct subspaces. This architecture allows non-critical dynamics, critical oscillations, and avalanches to coexist simultaneously. Empirical validation was performed by analyzing multiplane calcium imaging records of large neural populations, covering nearly 10,000 cells in the mouse visual cortex. By applying Principal Component Analysis (PCA) instead of the population average, the team isolated low-dimensional components (such as PC1 and PC2) and revealed robust, scale-invariant critical fluctuations that were invisible in the original signal.

“We show that hidden criticality, as predicted by our model, is prominent in recordings of large neural populations in mouse visual cortex.”

Implications and the Future of Neural Analysis The results of this study indicate that criticality may be far more prevalent in the brain than previously estimated. The discovery raises the urgency to develop new analytical approaches capable of revealing signatures of criticality that the population average fails to capture, adequately dealing with mixed modes and anticorrelated neurons. Methods that avoid grouping competing units have the potential to transform our understanding of brain dynamics, correcting observation flaws and expanding the empirical evidence on the optimal regimes of neural computation.