Computational Psychiatry & Development

Contributed Talk Session: Thursday, August 6, 1:45 – 2:45 pm, Skirball Theater

Localizing excitation-inhibition imbalance in schizophrenia with virtual brains informed by white-matter microstructure

Talk 1, 1:45 pm

Kailin Zhu¹, Georg Reich², Xiaojun Zhou³, Trang-Anh Estelle Nghiem³; ¹Hertie Institute for AI in Brain Health, ²Technical University of Berlin, ³Eberhard-Karls-Universität Tübingen

Presenter: Kailin Zhu

Schizophrenia symptoms are believed to emerge from cellular-level excitation-inhibition imbalance. Yet, where imbalance may be localized in the brain to potentially cause brain dynamical differences underlying altered cognition in each patient remains unclear. To investigate cellular mechanisms underlying brain function and dysfunction at the individual level, personalized whole-brain models provide promising tools. However, their applicability to psychiatry is still limited as models fail to account for inter-individual differences in brain dynamics. We hypothesize that models can be substantially enhanced by incorporating any information about white matter microstructure. Here, we systematically compare metrics of white matter structure and microstructure to inform personalized simulations of brain activity in schizophrenia and controls. To do so, we infer regional parameters of whole-brain mean-field models with The Virtual Brain (TVB) to account for individual functional connectivity (FC) from resting-state functional magnetic resonance imaging (fMRI) data. Our results show that models informed by white-matter microstructure metrics drastically outperform models informed by white-matter fiber count and density as in the state of the art at reproducing empirical FC. The findings support that the regional specificity, but not individual specificity of white-matter microstructure metrics, influence data-model fit. Next, we reveal E/I imbalance localized in the posterior cingulate and paracentral areas in schizophrenia patients. Finally, we demonstrate that inferred E/I maps are meaningful in that they can enhance diagnostic classification. Our approach provides a white-matter-microstructure-informed platform to model brain activity at the individual level in health and pathology, allowing us to introduce and validate tools to map E/I balance, supporting machine-learning-based diagnostics and potentially treatment simulation for personalized intervention recommendations.

Distinct Computational Pathways to Persistent Fear Across Development

Talk 2, 1:55 pm

Camilla van Geen¹, Noam Goldway², Gili Karni¹, Isabel M Berwian¹, Catherine Hartley³, Yael Niv¹; ¹Princeton University, ²King's College London, ³New York University

Presenter: Camilla van Geen

Exposure-based therapies often fail to produce lasting fear extinction. Computational models of latent cause inference highlight that persistent fear may arise when inferred aversive causes are selectively maintained in memory. Here, we extend this framework to include threat generalization as another pathway to sustained fear, and test whether the balance between selective maintenance and generalization shifts across development. Participants (N=502) aged 8–25 completed an online fear conditioning and extinction task, followed by a test of spontaneous recovery. Participants learned to associate an image (CS⁺) with an aversive scream, while a second image (CS⁻) was never followed by the scream. Compared to adolescents and adults, a larger proportion of children showed fear generalization across the two CSs, and a smaller proportion exhibited CS⁺⁻selective fear at test (both p_FDR = 0.012). Our model captured this developmental shift as an age-related decrease in a generalization parameter that spreads learning across multiple latent causes (i.e., associations; p = 0.037), alongside faster decay of existing memory traces (p < 0.001). These findings suggest a developmental transition from diffuse threat generalization to more selective maintenance of aversive memories.

The developmental trajectory of temporal dynamics in human visual representations

Talk 3, 2:05 pm

Chun-Hui Li¹, Marlena Baldauf¹, Siying Xie¹, Teresa Sylvester¹, Christina Maria Schaetz¹, Ali Mohammad Nezhad¹, Bert Turtleton¹, Stefanie Höhl², Moritz Köster³, Radoslaw Martin Cichy¹; ¹Freie Universität Berlin, ²Universität Wien, ³Universität Regensburg

Presenter: Chun-Hui Li

Visual representations must support both robust and rapid object processing, yet how these representations develop remains unclear. Here, we used electroencephalography (EEG) and multivariate analyses to characterize visual representations emerging during passive viewing of everyday objects, from infancy to adulthood (total N=190) from 0.5 years to adulthood in a cross-sectional study. We observed that the peak latency of object decoding decreased with age, indicating faster processing. Time generalization analysis revealed increasingly persistent representations with age. Together our results provide a first comprehensive cross-sectional view of the neural dynamics with which object representations emerge during development.

Self-Supervised Modelling of Social Primitives in Human Dyads

Talk 4, 2:15 pm

Thibaut Chataing¹, Thomas Maillart², Nada Kojovic², Giuseppe Chindemi², Sara Seyed Akhavan¹, Camilla Bellone¹, Marie Schaer¹; ¹University of Geneva, ²ETH Zurich

Presenter: Thibaut Chataing

Social interactions rely on dyadic primitives that are central to brain social perception. Yet they have rarely been explicitly modeled to computationally. We introduce HumanLISBET, a self-supervised pose model that encodes clinician-child dyads. Using 65 hours of unlabeled interactions for training, we evaluate the model on 119 children across 19 clinical targets (ADOS-2, Vineland-II). We evaluate HumanLISBET against several benchmarks, including a kinematic null model. Our model proves competitive for autism diagnosis (AUROC=0.77) and significantly outperforms baselines on relational Vineland-II subscales (Personal, Community), though it underperforms on ADOS-2 severity and the non-relational Domestic subscale. This selective pattern suggests that the embedding captures relational social dynamics, while session-level aggregation discards part of the temporal structure needed for social affect.

Vision Models Capture Complementary Representations in Children That Converge in Adults

Talk 5, 2:25 pm

Domenic Bersch¹, Timothy Schaumlöffel¹, Michela Proietti¹, Antonia Franaszek-Traczewska², Hannah Elisabeth Zwad², Siying Xie², Marlena Baldauf³, Teresa Sylvester², Christina Maria Schätz⁴, Moritz Köster³, Stefanie Höhl⁴, Bert Turtleton², Radoslaw Martin Cichy², Gemma Roig¹; ¹Johann Wolfgang Goethe Universität Frankfurt am Main, ²Freie Universität Berlin, ³University of Regensburg, ⁴University of Vienna

Presenter: Domenic Bersch

How the relationship between image-text aligned and vision-only models changes across development remains unknown. To address this, we applied time-resolved representational similarity analysis to EEG data from 8-year-olds, 12-year-olds, and adults. Commonality analysis showed complementary structure for image-text aligned versus vision-only models at 8 years, near-zero shared variance at 12 years, and greater overlap in adults, while both retained unique variance at all ages. This pattern generalized across model pairings, whereas vision-only model pairs showed overlapping variance at every age, suggesting a double dissociation. Within-category structure appeared complementary even in adults, suggesting that image-text aligned training may capture fine-grained structure not recovered by vision-only training. Together, these findings suggest that image-text aligned and vision-only models capture distinct aspects of visual representations that become more aligned in adulthood, and that vision-only models alone may miss structure relevant for developmental model-brain alignment.

Do you see what I see? Mobile eye-tracking in natural environments reveals selective social-visual engagement and developing visual field biases in children and teens

Talk 6, 2:35 pm

Elizabeth Jiwon Im¹, Kalanit Grill-Spector¹; ¹Stanford University

Presenter: Elizabeth Jiwon Im

Face recognition and word reading improve during childhood and adolescence alongside increasing cortical selectivity for faces and words, while selectivity for hands decreases. These developments may be linked to changes in visual diet. However, the naturalistic visual diet of children and teenagers is unknown. For the first time, we use mobile eye-tracking to compare the visual diet and gaze behavior of children (4–6 yrs) and teenagers (14–17 yrs) in home and classroom environments. We find that the prevalence of social-visual information in children's and teenagers' visual diet does not always predict how they engage with that information. Moreover, different contexts lead to different visual statistics and gaze behaviors, even within the same age group. We also find that visual-field biases continue to develop from childhood into adolescence. Together, our findings suggest that children and teenagers sample the world differently, with context shaping gaze behavior across development, suggesting that models of visual development should incorporate not only scene statistics but also age- and context-dependent fixation statistics.