Abstract
Brain areas are organized into functionally connected networks though the mechanism underlying this widespread coordination remains unclear. Here we apply deep learning-based dimensionality reduction to task-free functional magnetic resonance images to discover the principal latent dimensions of human brain functional activity. We find that each dimension corresponds to a distinct and stable spatial activity gradient. Brain areas are distributed non-uniformly along each gradient, reflecting modular boundaries and hub properties. Gradients appear to dynamically steepen or flatten to produce task-specific activation patterns. Dynamical systems modelling reveals that gradients can interact via state-specific coupling parameters, allowing accurate forecasts and simulations of brain activity during different tasks. Together, these findings indicate that a small set of overlapping global activity gradients determine the repertoire of possible functional connectivity states.
Competing Interest Statement
The authors have declared no competing interest.