ASD-DyadHyperscan: A multimodal fNIRS hyperscanning, behavioral, and linguistic dataset of parent–child interactions in autism

Investigating interpersonal neural coordination and communicative attunement during real-world social interaction is essential for unraveling the neurobiological mechanisms underlying social impairments in autism spectrum disorder (ASD). However, open-access hyperscanning datasets involving clinical pediatric populations remain virtually non-existent. Here, we present the ASD-DyadHyperscan Dataset, a comprehensive multimodal benchmark comprising synchronized recordings from 92 child–caregiver dyads (44 autistic children and 48 typically developing children matched for age and sex, alongside their primary caregivers). The dataset captures dyadic interactions across three sequential naturalistic contexts: cooperative puzzle solving (5 minutes), unstructured free talking (5 minutes), and a post-task resting baseline (3 minutes). Data records integrate: (i) continuous dual-brain fNIRS hemodynamic time series spanning 16 fronto-temporal regions of interest; (ii) micro-coded video behavioral annotations derived via the Dyadic Parent-Child Interaction Coding System (DPICS); (iii) digitized conversational transcripts and quantitative linguistic metrics computed using Computerized Language Analysis (CLAN); and (iv) comprehensive clinical, cognitive, adaptive, executive, and emotional phenotyping (ADOS-2, WISC-IV, Vineland-2, BRIEF-2, RCADS). Technical validation demonstrates robust optical coupling and physiological pulse signal-to-noise ratios, established inter-rater reliability, distinct behavioral and syntactic group profiles, and strong cognitive-adaptive covariation. Curated under Brain Imaging Data Structure (BIDS) guidelines, this dataset provides a standardized resource for computational psychiatry, inter-brain synchrony modeling, and dyadic neurodevelopmental research.