Network Neuroscience

The computational study of epileptic seizure dynamics has primarily focused on the identification of seizure onset zones and propagation pathways. Here, we present a network dynamical model implemented on the empirically measured mesoscale mouse brain network that reveals previously unresolved organizational principles underlying seizure propagation. Rather than the conventional assumption of a single dominant pathway to synchronization, the model reveals multiple competing pathways to synchronization with distinct dynamical properties including transition propensity, recruitment speed, spatial coverage, synchronization stability and transition kinetics. Consequently, node ablation does not uniformly suppress synchronization across pathways, but instead selectively alters pathway occupancy, producing non-trivial alterations to seizure dynamics with potential implications for resection and network-targeted intervention strategies. Biologically, the olfactory and limbic sub-networks emerge as key mesoscale regulators of synchronization dynamics, with olfactory recruitment preferentially constraining global synchronization while limbic-driven pathways preferentially support seizure generalization. More broadly, these findings extend transient explosive synchronization theory by demonstrating that synchronization in biologically constrained networks may emerge through competing mesoscale recruitment programs rather than a single transition process. Together, these findings introduce a new conceptual framework for seizure propagation, suggesting that pathological synchronization emerges not through a single dominant route, but through competing mesoscale dynamical pathways whose accessibility depends on both network architecture and ongoing network state. Significance statementEpileptic seizure propagation is conventionally understood as progressing through a dominant pathway that recruits increasingly larger portions of the brain into pathological synchronization. Using a network dynamical model implemented on the empirical mesoscale mouse connectome, we show that seizure-like synchronization instead emerges through multiple competing pathways with distinct spatial and temporal characteristics. These pathways differ in their propensity for generalization, synchronization stability and sensitivity to node perturbation, such that network interventions selectively reshape pathway accessibility rather than uniformly suppressing seizure dynamics. Our findings introduce a new framework for understanding seizure propagation, identify mesoscale mechanisms linking network architecture to synchronization dynamics and suggest that competing synchronization pathways may represent an important organizing principle in complex brain networks. Graphical abstract O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=97 SRC="FIGDIR/small/730931v1_ufig1.gif" ALT="Figure 1"> View larger version (36K): org.highwire.dtl.DTLVardef@7c4076org.highwire.dtl.DTLVardef@16c0961org.highwire.dtl.DTLVardef@1dbd8a0org.highwire.dtl.DTLVardef@6b06de_HPS_FORMAT_FIGEXP M_FIG C_FIG

Cerebral cortical networks in the mammalian brain exhibit a non-random organization in which reciprocal projections, although widespread, are systematically asymmetric in strength: feedforward connections are consistently stronger than their feedback counterparts, particularly in sensory cortices. This "no-strong-loops" principle is thought to prevent runaway excitation and maintain stability, yet its actual computational impact remains unclear. Here, we use computational analysis and modeling to show that connectivity asymmetry supports high working-memory capacity, whereas increasing reciprocity reduces memory capacity and representational diversity in reservoir-computing models of recurrent neural networks. We systematically examine synthetic architectures inspired by mammalian cortical connectivity and find that sparse, modular, and hierarchical networks achieve superior performance, relative to random, small-world, or core-periphery graphs, but only when reciprocity is constrained. Validated on directed mammalian (macaque, marmoset, rat, and mouse) connectomes, these results indicate that restricting reciprocal motifs yields functional benefits in sparse networks, consistent with an evolutionary strategy for stable, efficient information processing in the brain. These findings suggest a biologically-inspired design principle for artificial neural systems.

Disorders of consciousness (DoC) are associated with large scale alterations in brain dynamics, yet the structural factors that constrain these changes remain unclear. Here, we investigate how the topology of the structural connectome shapes the sensitivity of brain dynamics to perturbation using a whole brain computational model constrained by diffusion MRI derived connectivity. We systematically probed the effects of node removal and targeted modulation of local excitation/inhibition balance on dynamic functional connectivity, quantifying dynamical richness via transitions between recurrent connectivity states and jump length distributions in functional connectivity space. We show that a node's integration within the structural connectome, quantified using a spectral integration measure, strongly predicts its impact on global brain dynamics. Lesions to highly integrative hubs drive the system toward low complexity dynamical regimes resembling those observed in DoC, particularly posterior medial regions such as the precuneus and posterior cingulate cortex. Analogously, increasing excitability in these regions restores healthy like dynamics in silico. In contrast, perturbations to weakly integrated regions have limited global effects. These results demonstrate that generic features of structural connectivity constrain whole brain dynamical stability and help explain why damage to specific hubs disproportionately disrupts conscious brain activity.

A central challenge in neuroscience is understanding how the human brain is organised to support optimal functioning and adaptability. One approach to characterise complex brain dynamics is by artificially perturbing whole-brain models. Here, we asked whether whole-brain organisation under perturbation in major depressive disorder (MDD) changes after intervention with psilocybin and escitalopram. First, we built whole-brain models of pre- and post-treatment resting-state functional magnetic resonance imaging (fMRI) and obtained an initial generative effective connectivity (GEC) matrix for each individual. Then, we employed systematic and local artificial perturbations across intensities, re-optimised each model to create a response GEC (GECr), and assessed the extent of brain reorganisation by quantifying the brain network reconfiguration index (NRI). Our results showed that the global brain NRI increases with psilocybin and decreases with escitalopram. Across sessions and interventions, higher global NRI was related with localised perturbations in brain areas orchestrating the brain's hierarchical dynamics. Traditional approaches complemented our investigation. Our findings suggest distinct neural changes following each treatment for MDD. The increase in brain reorganisation under perturbation following psilocybin is consistent with greater brain flexibility and changeability, whereas the decrease following escitalopram suggests more stabilised brain dynamics. Overall, perturbation-induced brain NRI may represent a useful approach for uncovering neural changes following different interventions for depression.

The brain requires coordination among different regions to execute cognitive tasks, which may involve both positive- and negative-correlations. The topology of these correlations may indicate the mechanism underlying brain functioning in a given state. Here, we study changes in the functional connectomes (FCs) of both the positive and negative-correlations across various cognitive task states relative to the resting state, using publicly available electroencephalographic (EEG) data. Considering the EEG-specific topographical cortical regions as topographical modules (TMs), we find that the FC comprising positive correlations (G+) is modular. In contrast, networks of negative-correlations (G-) are anti-modular, with more connections between TMs than within them, and are associated with improved overall topological efficiency. These functional networks also show variability across frequency bands and brain states. In the low-frequency delta band, resting states exhibit higher modularity and anti-modularity than task states; in contrast, in the high-frequency Gamma band, modularity and anti-modularity are much higher during task states than in the resting state. The k-core analysis of all networks further reveals differences: G+ is more hierarchical and robust than G- across all states. Moreover, the task-state networks are always more hierarchical than the resting-state networks across all frequency bands. In the high-frequency gamma band, they are also significantly more robust than the resting-state networks. These networks also differ in the topology of their innermost core constituents: the innermost core regions of G+ are randomly connected and spatially localized, mostly in posterior brain regions across subjects, in the high-frequency gamma band. Whereas those in G- are spatially de-localized, cover the extreme anterior and extreme posterior brain regions, and remain anti-modular in all the frequency bands. Overall, our analysis reveals the presence of an anti-modular organization of functionally specialized TMs alongside their modular organization and points to task- and resting-state differences in their topologies.

Amyotrophic lateral sclerosis (ALS) is increasingly recognized as a multisystem neurodegenerative disorder in which motor-neuron degeneration is accompanied by widespread alterations in cortical dynamics. Among its most reproducible neurophysiological signatures is cortical hyperexcitability, yet how this local excitability imbalance shapes distributed whole-brain activity remains poorly understood. Here, we combined source-reconstructed resting-state MEG data, tractography-informed whole-brain modeling, and simulation-based inference to investigate whether ALS-related alterations in large-scale brain dynamics can be mechanistically explained by changes in cortical excitability. First, we characterized empirical brain dynamics using complementary features spanning regional activity amplitude and variability, functional connectivity, and avalanche-based metrics. These analyses revealed significant alterations in ALS patients relative to healthy controls, as well as associations with clinical impairment and disease staging. To mechanistically interpret these changes, we employed a reduced Wong-Wang whole-brain model in which local recurrent excitation modulates emergent large-scale neural dynamics. Simulations showed that increasing excitability systematically reproduced the empirical dynamical signatures observed in ALS. We then applied a simulation-based inference framework to estimate latent excitability parameters directly from empirical observations. Whole-brain model inversion revealed increased excitability in ALS patients compared with controls. The recovered excitability parameter was associated with disease staging, supporting its clinical relevance as a model-derived descriptor of ALS progression. Finally, by extending the model to estimate frontal and non-frontal excitability separately, we found that ALS-related alterations were predominantly associated with increased frontal excitability, whereas non-frontal regions appeared comparatively less affected. The recovered parameters related to disease staging. Together, these findings provide a mechanistic framework linking altered large-scale brain dynamics in ALS to selective cortical hyperexcitability, explaining how local excitability changes can give rise to global network reorganization. More broadly, they show how computational model inversion can recover latent multiscale pathophysiological processes from empirical neural recordings, offering a non-perturbative alternative to complex experimental paradigms typically required to causally probe local-to-global mechanisms.

Neurodegeneration progressively removes synapses and neurons, yet neural circuits can retain stable collective dynamics despite substantial structural loss. Which structural principles confer this resilience remained unclear. Using large-scale spiking networks spanning empirical and synthetic microcircuit architectures, we systematically compared synaptic and neuronal modes of degeneration under controlled pruning strategies. We found that resilience was not determined by connectivity loss alone, but by how inhibitory gain was embedded within circuit architecture. Networks in which inhibitory neurons occupied structurally central positions robustly maintained health-like firing rates, levels of synchrony, and informational bandwidth across degeneration stages, whereas architectures lacking such embedding exhibited amplified dynamical disruption. Across regimes, the evolution of activity was organized by a compact set of weight-aware structural descriptors that generalized across network sizes and classes, with total effective synaptic coupling providing a dominant organizing axis. These results identified inhibitory architecture as a mechanistic determinant of circuit resilience and provided a predictive framework linking structural degeneration to collective dynamics.

Structural connectivity (SC) data are crucial for brain network analysis, but SC-based machine learning often suffers from limited data availability, hindering model generalization and robustness. Although data augmentation using deep generative models has attracted increasing attention, it remains unclear how different models capture the complex topological features of SC data. To clarify the learning characteristics of deep generative models for SC generation, this study compares three representative models: variational autoencoder (VAE), Wasserstein GAN with gradient penalty (WGAN-GP), and denoising diffusion probabilistic models (DDPM). We systematically evaluated these models using both synthetic datasets with known characteristics and real-world SC data. Generation quality was assessed using graph-theoretic metric comparisons and visual inspection of the generated adjacency matrices. WGAN-GP showed relatively stable performance across datasets and metrics, without severe performance degradation across evaluation settings. In contrast, VAE and DDPM performed well in specific aspects but were more sensitive to data characteristics. These findings suggest that WGAN-GP may serve as the most balanced baseline for future SC data augmentation studies, whereas VAE and DDPM may be useful depending on the target application and structural properties of interest. Furthermore, because all models struggled to fully reproduce strict global constraints such as planarity, our results suggest that standard generative models may be insufficient to capture the complex topological features of SC data. This highlights the importance of incorporating the desired structural properties into the training or generation process.

Micro-electrode array (MEA) recordings are widely used to characterize functional connectivity in neural cultures and have gained traction for the analysis of human brain slices. However, the impact of graph construction methodology on the resulting network topology has not been systematically quantified. Here, we benchmark three methods - shared spiking activity, Pearson cross-correlation, and the spike time tiling coefficient (STTC) - across 37 recordings from human cortical slice cultures classified into low, moderate, and high activity groups. We show that method choice alone produces large topological differences (Cohens d = 0.86-1.14 for clustering coefficient, d > 1.0 for node count), while higher-order features such as modularity remain stable. Each method exhibits a distinct sensitivity profile: shared spiking detects activity-dependent changes primarily through network size, correlation uniquely captures clustering differences, and STTC combines strong biological sensitivity with negligible parameter dependence across lag windows (all d < 0.1). Within shared spiking, z-score normalization dominates all other parameter choices (d > 1.0 versus bin size effects of d < 0.23), functioning as an implicit analytical null model that fundamentally reshapes the edge set rather than merely rescaling weights. Inter-method edge overlap is low (Jaccard index 0.08-0.45) and activity dependent, demonstrating that these methods identify substantially different connections from identical data. Our results reveal that methodological choices including construction method, threshold, and normalization introduce hidden degrees of freedom with effect sizes comparable to the biological signals being measured. We provide practical recommendations for parameter selection, reporting, and cross-method validation in MEA-based network neuroscience. Author SummaryWhen we record electrical activity from brain tissue using grids of electrodes, we can ask how different sites influence one another and map the tissue as a network of connections. Thanks to novel culturing methods, this approach is increasingly used to study human brain slices. However, deciding what is "connected" is not well defined. Researchers use several different methods, and it has never been clear how much this choice shapes the network they end up describing. Here we compared three widely used methods on 37 recordings from human cortical slices spanning a range of activity levels. We found that the method alone can change the apparent structure of the network as much as real biological differences do. The methods frequently disagreed about which connections exist and some technical choices, including normalization techniques, had surprisingly large effects. Because these hidden choices can rival the biological signal, we provide this benchmarking work with practical recommendations for selecting, reporting, and cross-checking methods, so that network studies of brain tissue become more transparent, comparable, and reproducible.

The human structural connectome, most commonly derived from diffusion-weighted imaging (DWI) and tractography, provides a macroscopic description of whole-brain wiring and serves as the structural foundation of network neuroscience, large-scale brain simulations, and personalized digital brain twins. However, tractography-derived connectomes are fundamentally limited by their inability to distinguish afferent from efferent connections, yielding networks that are undirected and therefore blind to the hierarchical organization imposed by the directionality of anatomical connections. In this study, we introduce a directed human structural connectome (DHSC) by transferring tracer-derived projection patterns from macaque to human using cross-species connectivity blueprints. Topological analysis of the DHSC manifests biological plausibility, a small-world network organization, and a directionality-based hierarchy, which offer the hierarchical organization of human brain networks. In the context of brain dynamics, the introduction of directionality reshapes the propagation and persistence of sensory inputs. DHSC also best captures the empirical spatiotemporal dynamics of stimulus-evoked brain activity. The findings demonstrate that anatomical directionality is a critical determinant of large-scale brain organization and dynamics. This provides evidence that directed connectome may offer potential advantages in large-scale simulations of the human brain. The resulting DHSC, along with all related analyses and data are openly available.

Structural covariance networks (SCNs) represent spatial patterns of covariation in brain morphology, often as a network of connections between nodes representing correlations in grey matter volume or cor- tical thickness measured by magnetic resonance imaging (MRI). SCNs have been suggested to reveal differences in functional organisation that are reflected in coordinated alterations to brain structure, and often these differences are sought in graph-theoretic measures such as the degree of clustering, segregation into distinct modules, or the characteristic path length between nodes. A common practice is to calculate SCNs for groups of interest, and use permutation testing to determine if they are significantly different for the measure of interest. However, the statistical validity of group comparisons using SCN-derived graph measures remains poorly understood. Here, we systematically evaluate the reliability of SCN estimation and downstream graph-theoretic anal- yses using structural MRI data from the Human Connectome Project ( = 1,096). We use simulations to show the effects of sample size and atlas dimensionality on SCN reliability. Using bootstrapping to characterise the distribution of SCN graph measures, we establish that small sample sizes systematically bias graph-theoretic measures including clustering, characteristic path length and modularity. Finally, we use simulations based on extrema from the bootstrapping distribution to characterise the statistical power and false discovery rate (FDR) for graph-theoretic between-group comparisons of SCNs, showing that at small sample sizes ( [&le;] 30) permutation testing is no better than chance. These findings suggest that many significant SCN group differences, particularly those using small sam- ples and high-dimensional parcellations, may reflect sampling noise rather than true biological differences. We recommend that future SCN studies use larger samples, coarser parcellations, and explicitly evaluate reliability before interpreting group differences.

The brain is the most captivating chef doeuvre of nature. Naturally then, the mind wonders about the process that births such a fascinating organ. Neurodevelopment is a complex yet robust phenomenon that conceals answers to our questions in its intricacies. In an attempt to shed some light on this matter, we study the developing brain connectome of the nematode, C. elegans across the post-embryonic phase. A tiny organism with only around 200 neurons comprising its brain and yet a diverse array of behaviors to display, it makes for a great model. Starting with most of its head neurons already present at hatching, the worm brain accumulates numerous more synaptic connections increasing the edge density. It maintains a weak connectivity throughout thereby, balancing global communication as well as hierarchy. At the mesoscopic level, we find that the core has a conserved backbone of persistent neurons along with a dynamic component formed of transient/recurring neurons. Moreover, the connectome has a rich club organization since the early stage which selectively strengthens indicating progressively denser connectivity among the integrators due to the previously reported asymmetric synapse addition. This asymmetry also shows up in the preservation of input hubs across development and the progressively more centralized organization of the in-degree k-core. Our work provides a new perspective into the neurodevelopment of the brain that may facilitate our understanding of its functioning.

Brain activity can be understood as a sequence of neuronal avalanches, i.e., transient episodes of coordinated activation that emerge across scales, from individual neurons and local networks to whole-brain dynamics. Avalanches are typically characterized by features such as size, duration, number of active components, and the silent time separating consecutive events. Although these features have been extensively characterized through their marginal distributions, their temporal organization and dependence on the underlying brain architecture remain poorly understood, leaving us without a framework for embedding fast neuronal avalanches within slower brain dynamics. Here, we analyzed eyes-closed resting-state magnetoencephalography recordings and the corresponding structural connectomes from 30 healthy participants to investigate the dynamics of avalanche sizes. We found that large avalanches preferentially followed short silent times, whereas small avalanches were more likely to occur after long silent periods. Based on the empirical joint distributions of avalanche size and silent time, we could define four types of events occurring above chance levels (avalanche large or small, preceding pause long or short). Mixed categories--combining a small value of one feature with a large value of the other--occurred more frequently than expected, while same-category events happened less often than chance. Furthermore, consecutive events tended to remain in the same category, a phenomenon referred to as persistence. We next investigated whether a brain regions connectivity profile shapes its propensity to participate in avalanches of different sizes. More strongly connected regions participated most often in small avalanches, whereas weakly connected regions were preferentially recruited during large avalanches. This pattern may reflect the greater sensitivity of highly connected hubs to fluctuations propagating through the network, resulting in frequent but spatially contained events. By contrast, the recruitment of more peripheral regions may require broader and stronger collective activity, occurring only during rarer, large-scale avalanches. In contrast, regional participation showed no clear association with the silent time preceding an avalanche. Together, these findings show that neuronal avalanches are neither temporally independent nor anatomically unconstrained: their sequence retains a memory of preceding events, while structural topology shapes which regions are recruited as avalanches grow. By connecting fast avalanche dynamics with slower temporal organization and the structural connectome, our results provide a multiscale framework for understanding how transient events are embedded within ongoing brain activity.

A major challenge in systems neuroscience is understanding how external perturbations interact with ongoing brain activity. Transcranial magnetic stimulation (TMS), increasingly used in both basic and clinical neuroscience and often combined with electroencephalography (EEG), provides a unique opportunity to probe this interaction. However, how intrinsic dynamics constrain the propagation of TMS-evoked activity remains poorly understood. In particular, effective connectivity (EC)--capturing directed, state-dependent interactions between brain regions--is thought to critically shape perturbational spread, yet remains difficult to estimate at the whole-brain EEG level. Here we introduce an analytically tractable, generative whole-brain model that links spontaneous EEG activity to cortical responses under perturbation. By deriving a closed-form expression for the models cross-spectral density, we directly fit empirical resting-state EEG spectra and infer biophysically interpretable local dynamical parameters without time-domain simulations. We then estimate stimulation-site-specific EC using only a small fraction of the TMS-EEG trials. The resulting model accurately predicts the spatiotemporal structure of TMS-evoked potentials (TEPs) in unseen trials. Moreover, even without subject-specific refitting, group-level EC templates capture canonical site-specific propagation motifs underlying single-subject early TMS responses. Together, our results establish an analytical framework for individualized whole-brain modeling of TMS-EEG with potential applicability to model-based neuromodulation.

Objective: Brain network controllability provides a framework for understanding how structural organization shapes brain dynamics, yet current models mainly rely on white-matter connectivity and may overlook the contribution of gray-matter architecture. Approach: We constructed a fusion network combining diffusion tensor imaging-derived white-matter connectivity with gray-matter morphological similarity and investigated its controllability, biological associations, heritability, phenotype prediction, and control energy. Main results: Controllability derived from the fusion network preserved key topological properties of the white-matter network and was associated with neurotransmitter systems and cerebral metabolism. Compared with the white-matter connectivity-based network, fusion-based controllability showed a systematic shift toward higher heritability, improved prediction of several individual characteristics and cognitive functions, and lower modeled control energy for activating resting-state networks. Significance: These findings suggest that incorporating gray-matter morphological information into a DTI-supported network provides a complementary structural representation for studying brain network controllability and state transitions. The lower control energy represents a model-derived transition cost and should not be interpreted as a direct measure of physiological energy expenditure.

Neural activity is organized in low-dimensional structure, yet functional connectivity in fMRI typically represents each brain parcel by a single voxel-averaged time series. This scalar representation makes whole-brain connectivity tractable but discards potentially informative dimensions of within-parcel BOLD activity. Here, we represent each parcel by a low-dimensional temporal subspace derived from its principal-component time series and use the RV coefficient to quantify connectivity between regional subspaces. Across Human Connectome Project resting-state and working-memory data, progressively expanding these subspaces reveals reproducible connectivity regimes with distinct network and identifiability profiles. At rest, connectivity constructed from the first principal component identifies individuals more strongly than either voxel-averaged functional connectivity or higher-dimensional subspace representations. During working memory, identifiability instead peaks after secondary components are included, indicating that the distribution of individual-specific information across the regional PCA spectrum depends on cognitive state. These patterns replicate across independent samples and remain robust across multiple parcellation resolutions. Together, our findings show that within-parcel BOLD structure contains identity- and statedependent information that is obscured by scalar regional summaries. Functional connectivity may therefore be better understood as a family of related connectomes indexed by the regional subspace retained, providing a general framework for mapping interactions between low-dimensional neural representations.

Volumetric electron microscopy (EM) has become a critical approach to generating high-resolution reconstructions of brain tissue. As the size of EM volumes increase, use of automated image segmentation within the reconstruction pipeline has become essential, although it generates errors that need correction. The proofreading and correcting of these errors has since become the dominant cost driver in the pipeline, but precisely estimating the sufficient number of proofreading edits to enable meaningful scientific analyses of the reconstructed neuronal networks remains a challenge. We present a fast, computationally inexpensive way to estimate the progress of a connectomic proofreading effort without requiring a priori knowledge of ground truth. We show that simple global graph invariants converge predictably to asymptotic limits with increasing numbers of proofreading edits, informing a quantitative "pencils down" criterion for proofreading completeness. We illustrate our method on two datasets in different stages of proofreading progress, a zebrafish spinal cord and the hemibrain Drosophila melanogaster dataset. Our method reduces the uncertainty associated with the planning and prioritization of proofreading activities and enables data owners to accurately predict and budget the amount of proofreading necessary for their scientific questions.

Brain oscillations emerge from nonlinear interactions across anatomically connected neural populations, alternating between transiently coordinated and desynchronised states. Transcranial alternating current stimulation (tACS) can modulate these dynamics, but most work has focused on frequency and amplitude, leaving waveform shape comparatively unexplored. Here we used a whole-brain model of delay-coupled Stuart-Landau oscillators constrained by empirical human structural connectivity to determine how sinusoidal, square, triangular, sawtooth and pulsed stimulation reshape spontaneous alpha-band activity. All waveforms were applied at the same frequency and amplitude to the posterior parieto-occipital regions. Network responses were quantified using the Kuramoto order parameter, spectral entropy and metastable oscillatory modes of transient alpha bursts. Sinusoidal and pulsed stimulation produced the strongest effects, increasing global synchrony while reducing metastability and spectral entropy, consistent with a transition from a fluctuation-rich regime to a coherent and spectrally concentrated state. These waveforms also transformed intermittent alpha bursts into prolonged or near-continuous oscillatory episodes. In contrast, square, triangular and sawtooth stimulation reduced synchrony while largely preserving metastability, producing weaker and more fragmented modulation. These findings identify waveform shape as a key determinant of rhythmic stimulation effects and a principled parameter for neuromodulation design.

Brain function is deeply embedded within multiscale structural architecture. Conventional studies predominantly utilize pairwise connectivity networks to investigate structure-function relationships. However, this low-dimensional perspective overlooks multi-region collaborations for complex cognition. Consequently, whether and how anatomy constrains such higher-order functional networks remains unresolved. To address this pivotal question, we utilize an information-theoretic O-information approach to characterize higher-order functional interactions (HOIs). By reconstructing individual-level HOIs from multimodal structural networks, we directly validate the structural constraint on HOIs. The resulting reconstruction coefficients are defined as structural-functional constraint strength (SFCS), serving as a quantitative vehicle to decipher how anatomy shapes these higher-order networks. SFCS uncovers a highly heterogeneous structural constraint landscape across data modalities, spatial regions, and informational interaction modes. Crucially, individualized SFCS robustly predicts multi-domain cognitive phenotypes, showing higher sensitivity for informant-reported than patient-reported assessments. Finally, we show that this landscape undergoes pathological, mode-specific reorganization in Alzheimers disease. Cross-scale alignment with spatial transcriptomics further demonstrates that this macroscale network remodeling is coupled with microscale metabolic and regulatory gene pathways. Collectively, our findings not only validate the structural constraint on higher-order functional networks but also decipher its precise underlying mechanisms. This constraint paradigm plays a pivotal role in shaping diverse cognitive capabilities, while its pathological disruption in Alzheimers disease highlights the potential of SFCS as a biomarker for tracking neurodegenerative network impairments.

Understanding dynamic mechanisms underlying cognition remains a major challenge in human neuroscience. Here, we develop, validate, and apply Multivariate Dynamical Systems Identification with Amortized Variational Inference (MDSI-AVI), a novel computational framework designed to address critical challenges in capturing asymmetric, context-dependent, whole-brain directed interactions while accounting for regional hemodynamic response variability in fMRI data. MDSI-AVI leverages simulation-based inference through forward and reverse variational inference to address the limitations of conventional variational methods in high-dimensional settings. By averaging over uncertainty in hemodynamic response parameters using forward simulation, MDSI-AVI provides well-calibrated posteriors of directed connectivity that scale efficiently to networks with hundreds of nodes. Applied to Human Connectome Project data (N=728), MDSI-AVI reveals new insights into working memory mechanisms, identifying the dorsal anterior insula as a critical hub influencing activity at the whole-brain level. We demonstrate task-dependent modulation of causal influences, where the salience network drives frontoparietal network activity, which differentially influences the default mode and sensorimotor networks depending on working memory load. These whole-brain causal interactions distinguish task conditions with high accuracy and predict working memory performance. Our framework demonstrates reproducible results across whole-brain parcellations, establishing MDSI-AVI as a robust tool for advancing our understanding of circuit dynamics in cognition and disease.