Fitting the Sherlock dataset using the pairwise maximum entropy model.

(A) The cortical ROIs were classified into seven functionally distinct brain networks, with network activity binarized using the network-averaged signal. The black line shows the mean activity across all seven networks. (B) Seven brain states were identified using the pairwise maximum entropy model; active networks are highlighted in blue, and each network activation pattern corresponds to a local energy minimum for that state. (C) Group-averaged transition probabilities and state occupancy are shown. Self-state transitions were excluded from the transition matrix and are left uncolored. (D) The relationships among the activity patterns corresponding to energy local minima are illustrated in a disconnectivity graph.

Route-related BOLD activity.

(A) Example energy landscape illustrating the route of attractor movement across three local minima. The transition from Tm-1 to Tm represents a hard route, characterized by a rapid increase in energy, while the transition from Tn-1 to Tn represents an easy route, marked by a sharp energy decrease. (B) Clusters demonstrating significantly increased BOLD activity following the transition, as identified by paired t-tests comparing post- and pre-transition BOLD values.

LASSO regression analysis of brain state dynamics and inter-subject correlation.

(A) ROI-wise ISC analysis in the Sherlock dataset across 114 selected ROIs. (B) 29 ROIs with significant R2 values after FDR correction within 114 ROIs. (C) Mean coefficient distribution across ROIs, with 26 out of 29 selecting S7→5 as a feature in all 17 leave-one-out LASSO models. (D) Hierarchical clustering of the filtered coefficient matrix (ROI × valid feature, 29 × 16), revealing three clusters at a distance threshold of 3.0. (E) Additional statistics derived from LASSO analysis show that clusters primarily consist of ROIs from the SAN, DMN, and subcortex, with varying frequencies of feature selection across models. (F) Linear regression between the transition probability and the values of ISC. Four typical ROIs were selected in three clusters.

Brain state transitions are accompanied by changes in thalamic activity.

Thalamic BOLD activity associated with eight state transitions was assessed across a 13-TR window centered on the transition onset (0 TR). The solid black line depicts the average BOLD response across participants, while the shaded gray band denotes the 95% confidence interval. The light gray region marks the upper and lower limits established through permutation testing. At each time point, independent t-tests compared the data from 17 subjects with 1,000 permutations; red stars indicate time points with significant differences (p < 0.001, Bonferroni-corrected for 13 comparisons).

Thalamocortical functional connectivity changes at brain state transitions.

Thalamocortical functional connectivity at brain state transitions was compared against null distributions generated by randomly placed transitions. Voxels with |z| > 1.95 (ps. < 0.025) are highlighted, with clusters corrected using AFNI (NN1, bi-side, 201 voxels). For S7→5, a more lenient threshold of |z| > 1.645 was applied to detect marginally significant clusters (ps. < 0.05) as we did not find any valid cluster at the original threshold.