Heterogeneous Encoding Patterns of Dynamic Audiovisual Social Stimuli in Autism Spectrum Disorder (Heterogeneous Encoding Patterns in Autism)
The neural basis of social difficulties in autism spectrum disorder (ASD) remains uncertain. Prior studies typically used static stimuli or reduced complex stimuli to only a few dimensions, obscuring real-world complexity. Here, we used videos with social content to engage brain networks and trained encoding models to predict brain activity from high-dimensional audio, visual, and more abstract socially relevant features (e.g., valence). We hypothesized that brain activity in autistic adults would be less accurately predicted and more heterogeneous within and across individuals than in neurotypical (NT) participants, particularly for socially relevant features. Participants [48 ASD (13 female), 90 NT (25 female)] viewed six videos (5–20 min). Brain activity was measured using fMRI. Cross-validated encoding models were assessed on prediction accuracy, variance, and pattern similarity. Brain activity was predicted significantly above chance across brain networks and videos with striking consistency between groups. Auditory and language networks were most accurately predicted. Within-participant predictions were equally variable between groups; however, autistic participants exhibited lower within-group encoding pattern similarity than NT participants and greater across-group than within-group similarity. Analyses using only socially relevant features showed the largest across-group differences in encoding pattern similarity. The novel application of encoding models to multiple videos in ASD suggests greater heterogeneity in neural representations of social information in ASD. Although group-averaged encoding accuracies were comparable across groups, ASD participants showed more heterogeneous regional encoding patterns. These idiosyncratic patterns underscore the need to incorporate individual variation in brain representations when developing translational neuroimaging approaches in autism. Significance statement Individuals with autism spectrum disorder (ASD) experience difficulties in social communication and interaction, yet the underlying neural mechanisms remain unclear. Here, we present a novel application of encoding models to neuroimaging in ASD. Neural representations of dynamic and complex social information were more heterogeneous across autistic as compared with neurotypical individuals while encoding model prediction accuracy was similar for population averages across groups. Effects were robust across six videos and spatial scales. Variability in neural representation across individuals, rather than systemic population differences, may thus be associated with atypical brain function in complex social information processing in autism. The findings highlight the importance of accounting for individual variability when studying the neurobiology of autism and developing translational neuroimaging approaches.
Authors
- Dorit Kliemann (ORCID: https://orcid.org/0000-0003-2069-919X)
- Lisa A. Byrge (ORCID: https://orcid.org/0000-0001-8554-1401)
- James A. Traer (ORCID: https://orcid.org/0009-0006-3262-9833)
- Daniel P. Kennedy (ORCID: https://orcid.org/0000-0002-5915-0893)
- Avery L. Van De Water (ORCID: https://orcid.org/0000-0003-3061-1964)
Publication Details
- Journal
- Journal of Neuroscience
- Published
- 2026-11-15
- DOI
- https://doi.org/10.1523/jneurosci.0055-26.2026
- Primary Topic
- Autism Spectrum Disorder Research
- Type
- article
- Field-Weighted Citation Impact
- 0.00