Network Neuroscience

Whole-brain connectomes are increasingly used as wiring diagrams for spiking simulations and as structural priors for artificial networks, usually after an ad hoc synapse-count threshold whose effect on validated function has never been measured. We define function-preserving sparsification - removing edges under a fixed budget while preserving validated behaviour and population responses - and study it in the leaky integrate-and-fire whole-brain model of Shiu et al. (2024). Six operators - weak-synapse pruning, uniform and weight-proportional random removal, degree-preserving rewiring, weight shuffling and a task-aware criterion - were compared at identical edge budgets across three validated circuits (sugar-evoked proboscis extension, its bitter suppression, grooming pathway specificity), silent and spontaneously active network states, and multiple seeds. Weak-synapse pruning preserves 79% of feeding output with 6% of edges, with a cliff at k* of approximately 9 synapses, whereas random, degree-preserving and weight-shuffled controls at the same budget abolish output, and weight-proportional random sampling survives only while it retains the strong-edge set: function resides in the conjunction of wiring placement and strength, not in degree statistics or synaptic mass. The cliff is task-dependent - grooming specificity collapses below k = 2 through disinhibition. In the active state failure changes character: output over-excites instead of vanishing, and bitter suppression is lost progressively from k = 5 and reverses at k = 31. A task-aware criterion preserves its training stimulus at 0.4% of edges but loses held-out inhibition first. These results provide a first benchmark for connectome sparsification and show that what must be kept depends on task and network state.

Lesion network mapping (LNM) has provided a compelling framework in clinical neuroscience, linking focal brain damage at heterogeneous anatomical locations to shared functional networks across behavioural syndromes. Recently, the methodological specificity of LNM has come under discussion following observations that derived network maps tend to converge onto the normative connectome's generic degree distribution. While current efforts largely treat this convergence as a statistical inference challenge addressable via post-hoc spatial null models or individual-level analyses, we identify a foundational upstream measurement artifact operating within individual patients: the lesion "bystander effect". Because pathological lesions follow biological constraints rather than functional brain parcellations, a typical lesion encapsulates both a symptom-driving critical node and surrounding asymptomatic bystander tissue. By averaging connectivity profiles across the entire lesion mask, LNM mechanically dilutes focal clinical signals. Across empirical clinical cohorts, individual-level machine learning classifiers, synthetic within-cohort lesion unions, and whole-brain spherical simulations, we demonstrate that increasing lesion extent deterministically drives network maps toward the connectome degree vector following a mathematical power law. Fulfilling the clinical promise of LNM will therefore require computational strategies that isolate the causal functional core from surrounding bystander tissue prior to network projection.

A brain's wiring must satisfy conflicting pressures, from the outside world (be capable) and from the inside (stay within finite resources). These pressures often have opposing requirements. Prior work studies these trade-offs mostly between 2-3 pressures at a time, and thus cannot say how many pressures shape brain wiring or for which trade-off brains decide. To see all possible trade-offs at once and to uncover the set of most important pressures, we generate 25,000 synthetic brain networks and describe each by 29 network properties. We then place 635 human (ages 6-22) and 224 mammalian (12 taxonomic orders) brain networks in this property space. The space is triangular, and follows three pressures: computational capability, robustness to damage, and metabolic efficiency. No network is optimal at all three, and biological brains cluster in one small niche of this triangle. There, they accept higher metabolic cost to gain both robustness and capability. Development and evolution move brains in the same direction; additionally, human maturation first raises capability, and later optimizes for efficiency. This study underscores the triangular trade-off between metabolic efficiency, computational capacity and robustness, and the narrow region brains occupy in this trade-off triangle, visible only from a multi-property, cross-species perspective.

A central goal of neuroscience is to establish how similar (or different) brain networks are across individuals, development, psychiatric conditions, and even between biological and artificial brains. Whenever one such comparison is made, it relies on the assumption that we have a measure that produces an accurate and plausible similarity score. Yet, the field lacks one such measure, and the wide variety of existing measures often produces conflicting results. To address this problem, we perform a systematic comparison of 16 established similarity measures. First, we show that different measures consistently disagree on which brain networks are most similar. Second, we rank all measures according to multiple criteria such as biological plausibility, computational efficiency, and sensitivity. We show that DeltaCon is the most accurate measure in terms of parameter recovery, and that it ranks among the fastest and most noise-tolerant. For this reason, we conclude that DeltaCon is the best general-purpose similarity measure. However, no measure ranks best on more than two of the five criteria we assessed, and different measures should be preferred depending on the exact research question. Measure selection can now be made on evidence and should be a crucial decision in any study that compares brain networks.

Brain network organization emerges from ongoing network dynamics that continuously create, maintain, and reorganize connectivity patterns. At both the structural and functional levels, these dynamics shape network evolution throughout the lifespan. Here, we introduce a data-driven framework based on Stochastic Actor-Oriented Models (SAOMs), which model longitudinal network changes as a process of local edge formation and dissolution driven by local-connectivity factors. Applied to functional and structural brain networks from an aging cohort, the resulting models recovered a set of interacting mechanisms that collectively promote hub formation, short-range connectivity, and modular organization. These factors were derived through a novel data-driven factor-selection procedure rather than specified a priori, allowing network mechanisms underlying longitudinal change to be identified directly from observations. Functional networks exhibited substantially higher rates of change than structural networks, suggesting more frequent opportunities for network maintenance and reorganization. Structural networks additionally showed despecification in the hub-forming mechanisms with cognitive impairment. These results support a view of brain network organization as an actively maintained process emerging from ongoing local connectivity dynamics. More broadly, this framework provides a general approach for studying longitudinal network dynamics through the estimation of network-generating mechanisms over time.

Why are some individuals able to recollect past events and imagine future events more richly than others? Here we examine how structural connectivity within a 'core network' supporting mnemonic cognition, substantially overlapping with the canonical default network, underpins inter-individual variation in autobiographical thought. We analysed state-of-the-art magnetic resonance imaging data from 109 adults collected on a high-gradient 3T Connectom scanner, in relation to their Survey of Autobiographical Memory (SAM) scores - a psychometric measure of self-reported episodic autobiographical, semantic, spatial, and future thinking abilities. We anatomically defined a core network, employed streamline tractography to derive participant-specific structural brain networks, and weighted their connections with the myelin water fraction derived from quantitative multicomponent relaxometry. We investigated associations between graph theoretical metrics of the core network and self-reported mnemonic abilities. We found a negative association between future thinking abilities and characteristic path length, and a positive association between future thinking abilities and mean clustering coefficient. Further analysis showed that the node-specific clustering coefficient of most core network nodes was positively correlated with future thinking scores, with frontotemporal nodes showing the strongest associations. Our findings indicate that both efficient integration and local segregation of the core network are important for trait-level future thinking.

Lesion Network Mapping (LNM) is a method used to map symptom-related circuits in the human brain. Recent work has raised concerns about the method and the specificity of LNM-derived networks, which often show high spatial overlap across unrelated conditions. Null models have been proposed to address the raised limitations1. Useful statistical tools in network neuroscience, they are reference procedures; in LNM, the normative connectome matrix C remains the source of the derived networks also under null modelling. Their broader role as a general solution for LNM therefore warrants careful evaluation. Here, we examine the proposed null models step-by-step and test them empirically across multiple datasets. We show that the proposed null-model correction effectively only replaces C with a transformed matrix C' (e.g., C' ~ C - mean(C)), one that is directly derived from C and retains much of its low-dimensional structure. Consequently, null-model-corrected LNM maps remain shaped by the same standard set of connectome-derived patterns and continue to show repetition and non-specificity across unrelated conditions. We discuss the scope and limitations of null models for LNM, showing that they can statistically benchmark LNMs, but that their application does not resolve the core limitation of non-specificity of LNM networks.

Researchers increasingly have access to neuroimaging data with repeated measures within subjects. Linear mixed-effects modeling offers a valuable way to analyze such data, but its application remains limited in the neuroimaging literature due to high computational cost and a lack of mainstream analysis packages. Here we benchmark a recently published algorithm, the Fast and Efficient Mixed-Effects Algorithm (FEMA), against a conventional implementation (MATLAB's fitlme, restricted maximum likelihood) across 94,830 functional connectivity edges. In contrast to the Adolescent Brain Cognitive Development (ABCD) Study Dataset used in the development and testing of FEMA, the dataset used in the current analysis has a smaller sample size (n = 378) and comprises individuals with chronic pelvic pain scanned at up to four visits over three years. Despite these differences, the two implementations produced highly concordant results: fixed-effect estimates and test statistics correlated at r [≥] 0.999 (Lin's concordance correlation coefficient [≥] 0.999), and the two approaches reached the same statistical conclusion for 99.98% of edges. FEMA completed each analysis roughly 20 times faster. Agreement was weaker for the estimated variance components, where FEMA attributed systematically less variance to the participant random effect, yielding a slightly lower intraclass correlation in approximately 82% of edges. We further show that most residual disagreement reflected FEMA's default variance-grid resolution rather than the estimator itself, and could be eliminated at negligible computational cost. These findings support the use of FEMA for connectome-wide analysis of repeated-measures neuroimaging data.

Estimating population-level dependence between brain regions is challenging when the number of regions greatly exceeds the number of individuals. We introduce an information-borrowing generalized Bayesian method for high-dimensional covariance estimation from region-level brain measurements. A divide-and-conquer parameterization separates marginal variances from correlation calibrations: an empirical \emph{pilot} correlation matrix--by default, obtained by standardizing a sample covariance matrix (SCM)--provides a reference structure that is adaptively calibrated through region-specific parameters shared across pairwise correlations, reducing the number of free parameters from $\mathcal{O}(d^2)$ to $\mathcal{O}(d)$ ($d$ is the number of regions) while guaranteeing a positive-definite covariance through the inclusion of a ridge term. Horseshoe priors on the calibration parameters enable selective borrowing and shrinkage of pilot information. Under suitable conditions on pilot accuracy, we establish theoretical guarantees for covariance estimation while retaining scalability in ultra-high-dimensional settings via a low-rank likelihood representation and a Woodbury-accelerated Hamiltonian Monte Carlo algorithm. Simulations showed lower estimation error than the SCM, bagging and iteratively refined principal orthogonal complement thresholding (POETRY). Applications to a semi-synthetic cortical-thickness and ADHD resting-state fMRI data revealed significant group differences in the default-mode--visual network and somatomotor network--limbic network blocks after Benjamini--Hochberg correction.

Drug-resistant focal epilepsy (DRFE) is understood as a complex brain network disorder. As a result, conventional pair-wise functional connectivity may fail to capture the higher-order interactions between groups of epileptogenic regions. We applied multivariate information theory to stereotactic electroencephalography-defined networks in retrospective resting-state fMRI analysis to determine whether the overall strength and dominant type of interdependencies between three or more regions, represented by the S-information and O-information respectively, could offer clinically relevant insights prior to surgical intervention in DRFE. We found networks involving the epileptogenic zone to be mostly dominated by re-dundant information-sharing, which was predicted by stronger structural and functional connections between regions. Non-temporal lobe epilepsy patients demonstrated greater redundancy and stronger comparative interdependencies than temporal lobe epilepsy patients, in line with a more complex network architecture. Among non-temporal lobe cases with postoperative follow-up, not seizure-free patients displayed more positive O-information deviation scores than seizure-free patients, suggesting elevated redundancy in networks associated with poorer surgical outcomes. Our findings potentiate the S-information and O-information as complementary higher-order descriptors of the epileptogenic network within personalised presurgical assessments.

Multi-echo functional magnetic resonance imaging (fMRI) acquires signals with distinct contrast profiles across echo times (TEs). Empirical evidence suggests that different TEs capture distinct signal contributions with varying sensitivity to BOLD and non-BOLD processes from long to short TEs (Stroman, 2002; Krishnamurthy, 2023; Dong, 2024). Previous literature has not yet examined how brain states extracted from temporal dynamic functional connectivity (dFNC) evolve across TEs. Here we present a methodological and empirical study of TE-dependent evolution of dynamic connectivity states, with a phenomenological two-component model for interpretation. Multiple echoes were acquired with an echo planar time-resolved imaging (EPTI) sequence. TE-specific components were extracted via atlas-based group ICA (Neuromark 1.0), and brain connectivity states were generated via windowed dFNC and spatial ICA clustering of windows. Within-subject and group-level states evolved across TEs in network organization and variance. States exhibited TE-dependent variance profiles, with distinct peak echo times; and group-level points of state transition were independently mapped to TE-dependent transitions in variance. We modified a bi-exponential model to simulate and model changes in the dFNC time-course across TEs. The bi-exponential model revealed high consistency with subject states (R2 fit=0.86, res. error=0.55, 86% model convergence rate), suggesting a two-component signal system contributing to dynamic TE-dependent connectivity changes in state. A significant correlation between relative model terms was estimated on a cell-wise basis across states (r(24) = -0.43, p=0.03, 95% CI [-0.70, -0.04]). This suggests that specific bi-exponential signal parameters (i.e., short FNC1,TE=0, long FNC2,TE=0, and long FNC2,decay) were associated with the observed TE-dependent variations in connectivity states across TEs. In sum, our analyses provide downstream implications for TE-aware state analysis dependent on connectivity magnitude (FNCTE=0) and rate of connectivity decay (FNCdecay), in a two-component signal model, with future implications for TE-dependent state biomarkers.

Alterations in impulsivity characterize several neuropsychiatric disorders, however the neural processes that underlie an impulsive choice have not been identified. A recent body of literature suggests that there may be generalizable aspects of neural representations across animals. Whether this is also true for dynamic rather than just static (like geometrical) properties of neural representations, and specifically in the context of impulsive behaviors, is less clear. Therefore, the goal of this study is to characterize generalizable features of latent dynamics that disambiguate an impulsive from non-impulsive choice across many different data sets and animals. A deep learning-based latent factor model was constructed from neural recordings that were acquired in rat anterior cingulate cortex while performing a delay discounting task, which is commonly used to assess impulsivity. Prior work shows this brain region is critical for the performance of this task. We leveraged piecewise linear recurrent neural networks (PLRNNs) to perform a nonlinear latent factor analysis across animals and experimental sessions. Our PLRNN explained a substantially larger percent of variance in fewer dimensions than linear approaches. We cross-validated performance of the model on test data and following the disruption of recurrent connectivity or auto-regressive latent dynamics. These tests confirmed that our fitting procedure consistently embedded generalizable neuronal dynamics within the PLRNN. Further, we found that the latent dynamics exhibited strong rotational features and progressed faster for impulsive choices. In summary, our model captures properties of the latent dynamics that are consistent across multiple data sets and disambiguate impulsive choices.

Whole brain tractography contains rich trajectory-level information about white matter organization, yet conventional structural connectomes capture only part of it. We introduce Structural Network Coupling (SNC), a network representation enabled by SILICA (Streamline Independent Component Analysis). SILICA decomposes tractography into structural source components and their streamline-specific loading profiles, while SNC characterizes relationships among these components through their shared trajectory expression. The resulting component-level network retains direct correspondence to the underlying white-matter pathways and supports network visualization and graph-based analysis. SNC produced anatomically interpretable coupling among trajectory-resolved white-matter components and revealed a highly organized group architecture that independently recapitulated commissural, projection, and association systems. It also showed strong hemispheric segregation and structured cross-system coupling, and was highly stable across split-half samples. SNC extends SILICA from trajectory decomposition to network-level organization and provides a flexible framework for future structural and multimodal network analysis.

Early brain development is intrinsically dynamic, yet most neonatal resting-state functional magnetic resonance imaging (rs-fMRI) studies rely on static and symmetric measures of functional connectivity. Here, we investigated whether response-based metrics can capture latent dynamical organization of the neonatal brain. We analysed rs-fMRI data from the Developing Human Connectome Project (n = 714). Using a statistical-physics framework grounded in the fluctuation dissipation theorem, we estimated intrinsic pairwise response functions between 46 cortical and subcortical grey matter regions, enabling quantification of regional responsiveness and response duration. Response-based measures differed markedly from conventional functional connectivity and predicted gestational age at birth from a single scan. Preterm infants showed reduced responsiveness and shorter response persistence compared with term born neonates. Importantly, directed analyses revealed a developmental reorganization of driver listener roles across the brain, with preterm infants preserving the regional ordering of the term hierarchy but expressing it at approximately half amplitude. These findings reveal that intrinsic responsiveness captures a previously underappreciated dynamical dimension of neonatal brain organization and provides sensitive markers of neurodevelopmental maturity. Response-based metrics may offer a principled framework for detecting early network dysmaturation following preterm birth. They further indicate that prematurity does not reorganize the directed hierarchy of the neonatal connectome so much as arrest its expansion.

The cerebral cortex is organized along a gradient of laminar complexity, indexed principally by the emergence and granularity of layer IV and by the size of layer III neuron somata, and cortical types are the ordered categories along it. The Structural Model builds on that gradient to predict laminar connectivity, plasticity and disease vulnerability (Barbas and Rempel-Clower, 1997; Garcia-Cabezas et al., 2019). Existing digital cortical-type maps are area-level: they assign one type per classical area and cannot represent a type boundary that runs through an area. We release cyto7, an open, vertex-level, seven-type cortical-type atlas delineated on the fsaverage surface following the Garcia-Cabezas protocol and revised by the protocol's authors. We distribute the atlas with a reusable quality-control and support framework. A topological audit reports the map against four spatial rules that follow from cortical type analysis and the dual origin of the neocortex theory: neighbouring vertices differ by at most one type, the least complex cortex forms one continuous limbic belt, eulaminate II forms a single matrix enclosing a few holes, and eulaminate III and koniocortex form isolated islands. These were imposed during construction rather than tested, and the audit reports what imposing them cost. A vertex-wise benchmark against the existing area-level map, one type per von Economo-Koskinas area, shows exact agreement at 52.5% of vertices, with disagreement almost always a single level and concentrated inside the large, heterogeneous areas that a single areal label cannot resolve. Where the two maps disagree, cyto7's label better matches the independent multimodal profile in 60.1% of pooled feature-by-vertex comparisons, against 46% under a rotation null and 54% under granularity-matched alternative partitions (p = 0.020), the harder of the two. A per-vertex anatomical support score, computed from anatomy alone, states how strongly each hand-drawn label is backed by the anatomical evidence, and is deliberately not a calibrated error probability. We then validate cyto7 against independent brain measures not used to define it, under a spatial-autocorrelation (spin) null with FDR control. Intracortical myelin (T1w/T2w) increases, cortical thickness decreases, and the principal functional-connectivity gradient reverses along the type axis (all q of about 0.006 or less). Source-reconstructed resting-state MEG dynamics also track the axis: the intrinsic timescale shortens toward koniocortex (rho of about -0.46) and survives the null, as do the spectral centroid and the slow/fast band-power ratio, whereas the aperiodic 1/f exponent does not. The laminar marker genes are where cyto7 differs most from a quantisation of a continuous map: it shows a larger absolute correlation with the layer-IV marker RORB (rho = +0.78) and the layer-V marker FEZF2 (-0.59) than seven-level septiles of myelin, of the functional gradient or of a histological gradient, on four of five markers, while tying or losing to those septiles on the macroscale proxies. These are larger observed associations against three tested quantisations, not a test that the differences between maps are reliable, and they do not establish the accuracy of any individual boundary. Defined from anatomy alone, cyto7 recovers the sensorimotor-to-association hierarchy and, on these markers, some of the laminar signal underlying it. It is released with its full pipeline and an uncalibrated anatomical support score as a reusable substrate for structure-function and modelling studies.

Astrocytes actively regulate synaptic transmission and neuronal excitability, yet their role in orchestrating macroscopic cortical network regimes and slow-wave oscillations remains an active area of reasearch. This study investigates how bidirectional neuron astrocyte interactions shape emergent population dynamics using a computational network model of excitatory and inhibitory neurons coupled to an astrocyte. The results identify astrocytic feedback topology, rather than astrocytic coupling strength alone, as a key determinant of emergent cortical network dynamics. By systematically dissecting pathway-specific connectivity, it has been shown that the neuronal population driving astrocytic activation and the neuronal population receiving gliotransmission jointly determine whether the network occupies asynchronous irregular (AI), synchronous irregular (SI), synchronous regular(SR), asynchronous regular(AR) or quiescent regimes.Directing gliotransmission selectively onto excitatory neurons consistently promotes population synchrony regardless of the population influencing astrocytic dynamics, whereas selective modulation of inhibitory interneurons induces network quiescence via strong suppression. Under dual-target gliotransmission, network synchrony is dictated by the population driving astrocytic dynamics: excitatory-only drive promotes synchrony, while combined or inhibitory-specific drive preserves asynchronous states. Furthermore, the model reveals that astrocytic signaling kinetics provide an additional temporal control mechanism that regulates the frequency and persistence of self sustained up states.

Population-graph models aggregate a query subject with an unlabeled cohort, so their reported performance can depend on the evaluation regime rather than on predictive information alone. Evaluation context has two separable axes: the visibility of the cohort during training, and the context available at inference. We study the training-time axis under a protocol fixed before analysis, on an ABIDE-I cohort of 871 subjects (20 acquisition sites). Two population-graph implementations are compared: a learned/adaptive graph (EV-GCN) and a static weighted graph (Parisot-GCN). Within each implementation the cohort, connectivity features, outer splits and training and selection rules are held fixed, and only the information available to the graph varies. The primary estimand is the implementation-by-training-context interaction: the difference between the P0-to-P1 AUC changes of the two implementations ({Delta}{Delta}AUC). Across three preselected outer splits (666, 1024, 2024) the direction is consistent and positive, with observed interaction range 0.06285-0.13090; uncertainty is assessed separately within each split by paired subject and site-cluster bootstrap resampling and is reported per split. Splits are replication units: they are not pooled and no cross-split average is reported, so the result is not a claim of invariance to arbitrary splits. These findings tie reported population-graph performance to the evaluation information regime and motivate explicit reporting of cohort visibility. The contribution is methodological: a two-axis taxonomy of information regimes, preselected cross-split evidence that the sensitivity of reported performance to cohort context depends on the implementation, and a reusable reporting checklist.

Quantifying neural health with a single measure remains a major unmet challenge. Here, we report a physiological marker based on the spatial distribution of signal-derived energy correlates in the resting brain. We hypothesize that energy organization in the healthy brain approaches an entropy-maximizing, homeostatically regulated state characterized by five properties: lognormal energy distributions, balanced spatial allocation, statistical independence between regions, low energy variance, and temporal stability. Together, these properties define an energetic profile that the Neural Health Index (NHI) summarizes as a single score, computable from (scalp/intracranial) electroencephalography, magnetoencephalography, or functional magnetic resonance imaging. Across 808 participants spanning neurological, psychiatric, neurodevelopmental, and neurodegenerative disorders, NHI distinguished healthy from clinical populations with sensitivity and specificity exceeding 95% in internal and external validation, scaled with disease severity, tracked pharmacological treatment exposure, and generalized to unseen diagnoses. These findings support NHI as a disease-agnostic, modality-invariant candidate biomarker for disease stratification and treatment monitoring.

Psychedelic compounds such as N,N-dimethyltryptamine (DMT) induce rapid and profound changes in conscious experience, requiring neuroimaging measures that capture how brain dynamics evolve across space and time. Emerging evidence suggests that rotational (vortex-like) dynamics may contribute to the coordination of large-scale neural activity. Here, we introduce two vorticity-based metrics; spiral vorticity (SV) and spatial spiral vorticity dispersion (SSVD), to quantify the rotational organization and spatial heterogeneity of large-scale cortical phase dynamics. Using fMRI data acquired under placebo and DMT in a within-subject design, we assessed these measures globally, across resting-state networks, and regionally to examine their relationship to subjective intensity. DMT increased SV and SSVD and reduced temporal irreversibility, indicating stronger rotational organization, greater spatial heterogeneity, and altered directional structure of cortical dynamics. Effects were non-uniform across functional systems, with the largest changes in default mode, visual, frontoparietal, and dorsal attention networks. Both vorticity measures tracked the temporal evolution of subjective intensity globally, while network-level analyses revealed spatially heterogeneous brain-experience coupling. Spatiotemporally resolved analyses further showed that the strongest reorganization of spiral dynamics emerged during offset and recovery rather than at peak subjective intensity. These findings identify spiral dynamics as a time-resolved neural signature of the DMT experience. By quantifying the rotational structure and spatial heterogeneity of cortical phase flow, vorticity-based measures provide a framework for linking large-scale brain dynamics to evolving subjective experience.

Resting-state connectivity can predict task-evoked fMRI activation, but correspondence with an individual task map may partly reflect a shared population pattern. We evaluated the Variational Resting-state-to-Task Prediction TransformeR (VRPTR), a three-dimensional encoder-decoder combining a compressed Transformer bottleneck, variational latent sampling, and multiscale skip connections. Models were trained in 360 healthy adults from the WU-Minn Human Connectome Project and evaluated in 40 held-out participants from the same dataset for the story-versus-math language contrast. VRPTR achieved mean voxel-map Pearson r=0.642 and Dice AUC=0.519, exceeding compact volumetric BrainSurfCNN-like and SWIFUN-like comparators by {Delta}r=0.0376 and 0.0335, respectively. A template fixed from the 360 training maps yielded residual correlation 0.364; VRPTR improved correlation but not mean absolute or root mean squared error over this template. A matched factorial ablation associated Transformer processing with improved map correlation ({Delta}r=0.0168); the variational main effect did not survive multiple-comparison correction. Component swaps indicated that subject-specific activation patterns depended more on multiscale skip features than on the bottleneck representation. Latent sampling produced a modest association between posterior standard deviation and voxelwise absolute error (mean Spearman {rho}=0.091). Raw 95% intervals covered only 4.7% of observed values. Five-fold calibration within the held-out cohort increased coverage to 94.8%, and spatially varying uncertainty modestly improved probabilistic scores over a constant-width model. Uncertainty also increased under input degradation. These findings support the contribution of Transformer processing to language-map prediction and distinguish it from subject-specific spatial conditioning and relative uncertainty estimation. Independent validation of calibrated uncertainty and evaluation in clinical cohorts remain necessary.