Individualized morphometric-similarity deviations in autism linked to cortical hierarchy and microarchitecture

Autism spectrum disorder (ASD) shows marked interindividual variation that complicates efforts to identify coherent neuroanatomical signatures. Here we combine subject-wise morphometric-similarity networks with hierarchical Bayesian normative regression to map per-region, per-subject departures from expected cortical organization across 308 parcels in a large multisite cohort. Normative z-maps reveal that ASD individuals have a higher prevalence of extreme, spatially focal deviations than controls, yet fewer than 12% of cases show extreme deviation in any single region. These deviations are not random: they preferentially redistribute morphometric similarity from association and heteromodal cortex toward sensory and visual territories and align with canonical cortical gradients. Network-level analyses identify a spatially localized subnetwork of altered similarity (dominant in occipito-temporal/visual nodes). Semi-supervised clustering of individual z-maps resolves a reproducible continuous axis of deviation whose opposite ends carry opposite global polarity and distinct cortical-molecular couplings revealed by transcriptomic and neurotransmitter-receptor density maps, as well as divergent cognitive meta-analytic associations. Together, these findings reframe ASD neuroanatomy as gradient-constrained and molecularly tethered at the individual level and support normative, subject-wise approaches for mechanism-informed stratification. Normative models of morphometric similarity across 1,344 brains map where each autistic individual’s cortex departs from expectation. Deviations are focal and largely idiosyncratic, yet shift similarity from association toward sensory cortex.