Distinct weak antisymmetric interactions shape human brain functions as probability fluxes

The functional computation of the human brain is hypothesized to arise from the collective behaviour of the underlying neural network. A method based on an idea from equilibrium statistical mechanics has been applied to infer the underlying interaction network among the brain regions from whole-brain imaging, but recent studies have challenged the equilibrium assumption. Without equilibrium, how does the human brain encode its function in biased state transitions? Here we show that the probability fluxes, which quantify the irreversibility of state transitions, exhibit unique, task-dependent patterns in spontaneous and task-induced whole-cerebral-cortex activity. We then fit an Ising model with asymmetric interactions and reveal that the symmetric interactions among the brain regions are strong and task-independent, whereas the antisymmetric interactions are weak and task-dependent, and that the fitted model reproduces the observed fluxes for most tasks. Our results indicate that the human brain performs its functional computation by subtly modifying the antisymmetric part of the interactions among the brain regions, offering a new explanation for how a similar interaction network supports diverse brain functions. The method can be applied to the time-series data of other high-dimensional many-body systems to reveal the probability fluxes and infer the underlying interactions among components.