A deformable attractor manifold organizes human resting-state brain dynamics

Authors: Anastasios P. Athanasiadis, Marmaduke M. Woodman, Amirhossein Esmaeili, Meysam Hashemi, Spase Petkoski, Viktor K. Jirsa

Presentation type: Poster at SNUFA 2026 online workshop (4-5 Nov 2026)

Abstract

Resting-state brain activity continuously reconfigures across distributed networks, yet how these transient configurations are organized within a low-dimensional dynamical landscape remains poorly understood. Here, we characterize the geometry and dynamics of the human resting-state manifold using time-resolved fMRI from 1,050 participants of the Human Connectome Project. We identified two distinct dynamical regimes based on global signal coherence: a temporally stable low-coherence state (LCS) and a transient high-coherence state (HCS). Although less frequent, HCS strongly tracked functional connectivity fluidity and exhibited a markedly richer repertoire of coactivation patterns. Low-dimensional projections revealed a unimodal stationary density for LCS, contrasting with a bimodal density for HCS, indicating a transition from predominantly monostable to bistable dynamics. Using Bayesian hierarchical modeling of network-specific stationary densities, we characterized these landscapes with quartic potential functions. Bimodality in HCS emerged predominantly from peripheral networks and increased along the variance-explained spectrum, where functional manifold geometry became increasingly associated with structural organization. The HCS further exhibited a hierarchy of intrinsic timescales, with the default mode network displaying the slowest dynamics and a near-critical, weakly stable potential. We then reconstructed the flow on the low-dimensional manifold and identified dynamically changing attractor structures across states. Transitions between HCS modes occurred predominantly through intermediate LCS-mediated switches, whereas direct jumps were comparatively rare and were frequently associated with high-amplitude cofluctuation events. These events also promoted extended exploration along the slow manifold. Finally, applying the resting-state state model to task fMRI revealed that HCS-like dynamics were enriched primarily during interleaved rest periods and that mode-specific occupancy was associated with fluid intelligence. Together, these findings suggest that resting-state dynamics are organized by a deformable attractor manifold, in which network-specific landscape geometry and transient cofluctuations jointly regulate the exploration of brain-wide configurations.

Are you the author and do you want to add some related links to this page? If so, check out our guide to adding related links and videos.