Izumiya, M. | Search bioRxiv

bioRxiv2026

Metastability in EEG phase synchronization networks is associated with autistic traits in a neurotypical cohort

Metastability is a fundamental dynamical property of large-scale brain networks and reflects the capacity of the brain to flexibly reorganize transient coordination patterns. In this study, we investigated whether metastable properties of resting-state electroencephalographic (EEG) phase synchronization networks are associated with individual differences in autistic traits. Resting-state EEG data from 88 neurotypical adults were analyzed using two complementary metrics: synchrony coalition entropy (SCE), which quantifies the diversity of transient phase synchronization patterns, and the metastability index (MSI), which quantifies temporal variance in global phase synchronization. SCE showed frequency-specific associations with the Autism-Spectrum Quotient (AQ) attention-switching subscore at 18-24 Hz and the communication subscore at 4-8 Hz, suggesting that frequency- and network-specific patterns of metastable synchronization are associated with distinct aspects of autistic traits. In contrast, MSI showed a modest association with the social-skill subscore in the lower-beta range, but this effect did not survive a cluster-based permutation test. This exploratory observation suggests that global synchronization variability may capture a weaker, complementary aspect of trait-related metastable dynamics. These findings suggest that, within a neurotypical population, individual differences in autistic traits may be more sensitively captured by the repertoire of transient phase synchronization patterns, as indexed by SCE, than by global phase synchronization variability, as indexed by MSI. Moreover, the associations of distinct AQ subscores with SCE in different frequency ranges suggest that different dimensions of autistic traits may be related to metastable network dynamics operating at different temporal scales. Author SummaryThe brain constantly coordinates activity across many regions, and this coordination changes over time rather than remaining constant. Understanding these dynamic patterns is important for explaining individual differences in cognition and behavior. In this study, we focused on a dynamical property called "metastability," which describes how brain activity flexibly shifts between different patterns of coordination. Instead of remaining in a stable state, the brain repeatedly forms and dissolves coordinated activity across regions. We analyzed brain signals recorded with resting-state electroencephalography (EEG) and examined whether these dynamic patterns were related to individual differences in autistic traits. We found that different aspects of time-varying coordination were linked to different dimensions of autistic traits in a neurotypical population. These findings suggest that examining how brain activity changes over time, rather than relying only on time-averaged measures, can reveal neural features associated with individual differences in autistic traits. Our study highlights metastability as a useful concept for understanding the flexible and dynamic nature of human brain function.

neuroscience↗

bioRxiv2026

Pretraining Objective Shapes Cross-Category Generalization in Affective Image Prediction: A Geometric Comparison of Vision Transformer Encoders

The geometry of representations learned by deep neural networks is shaped jointly by architecture and pretraining objective, yet disentangling these two factors remains difficult. Here we isolate the contribution of pretraining objective by comparing two Vision Transformers from the same backbone family but trained under different objectives: language-image contrastive learning (CLIP) and ImageNet-21k classification. Using continuous Valence-Arousal prediction on the OASIS dataset as a probe of representational quality, we evaluated frozen features under Leave-One-Theme-Out and Leave-One-Category-Out cross-validation, the latter requiring extrapolation to entirely unseen semantic categories. The contrastively pretrained encoder generalized substantially better than the classification-pretrained encoder under both protocols, with the gap widening sharply when held-out categories required cross-category generalization. To characterize why the two representations differ, we developed a geometric analysis of prediction errors, treating per-image errors as vectors in the affective plane and quantifying their spatial structure via weighted phase-locking, trajectory-based occupancy entropy, and effective dimensionality. The classification-pretrained representation collapsed errors into a small number of attractor regions with a strong center-ward pull, whereas the language-aligned representation distributed errors broadly across the affective space. Layer-wise linear probing further revealed that affective information was distributed across depth in the contrastive encoder but increasingly concentrated in deeper layers of the classification encoder, mirroring the texture-bias and category-anchored statistics characteristic of ImageNet-trained representations. These results provide a representation-geometric account of how the choice of pretraining objective, holding architecture constant, determines whether learned features generalize across semantic boundaries or remain confined to category-bound visual regularities. HighlightsO_LIIsolate the effect of pretraining objective by holding the Vision Transformer backbone constant. C_LIO_LIContrastively pretrained features generalize across unseen semantic categories where classification-pretrained features fail. C_LIO_LIIntroduce a geometric analysis of prediction errors based on phase-locking and occupancy entropy. C_LIO_LIClassification pretraining produces concentrated error attractors and a rigid centerward bias. C_LIO_LIAffective information is distributed across depth in CLIP but localized in late layers of the classification ViT. C_LI

neuroscience↗