Deep Learning of Brain-Behavior Dimensions Identifies Transdiagnostic Biotypes in Youth with ADHD and Anxiety Disorders

Attention-deficit/hyperactivity disorder and anxiety disorders are highly prevalent in youth and are characterized by substantial heterogeneity and frequent co-occurrence. This transdiagnostic complexity challenges conventional diagnostic frameworks that rely on symptom-based categories, which often obscure underlying dimensional and neurobiological mechanisms and offer limited neurobiological specificity. To address these issues, we developed a deep learning-based brain-behavior modeling framework that integrates clinically salient functional connectivity with cognitive and behavioral measures to identify interpretable dimensions and biologically grounded subtypes (biotypes). We applied our model to the Adolescent Brain Cognitive Development (ABCD) dataset comprising 3,508 children aged 9-11 years and revealed two reproducible brain-behavior dimensions that captured variation in cognitive control and emotion-attention regulation. These dimensions further yielded three distinct biotypes, each exhibiting unique symptom profiles and distinct brain development. We tested the robustness and generalizability of the dimensions and corresponding biotypes in an independent cohort of 224 age-matched participants from the Healthy Brain Network (HBN) and documented their early expression before symptom onset during adolescence. These findings highlight the utility of brain-behavior dimensions for elucidating heterogeneous psychiatric presentations and advance a biologically grounded framework for early classification and potential clinical translation in youth mental health.