Identifying Two Autism Biotypes Using Multi-Task Learning Derived Individual-Specific Functional Connectivity.

purposeAutism spectrum disorder (ASD) is a complex neurodevelopmental condition with considerable heterogeneity in clinical symptoms and underlying neurobiological mechanisms. This study aims to identify distinct ASD biotypes by leveraging healthy control (HC) information to extract individual-specific functional connectivity (ISFC) of ASD patients for clustering.

methodsThe study included resting-state fMRI data from 299 male ASD patients and 243 male healthy controls (HCs). We employed multitask learning–based sparse convex alternating structure optimization to extract group-shared and individual-specific connectivity patterns, followed by clustering analysis based on ASD patients’ ISFC features. Subsequently, we compared clinical symptoms, functional connectivity differences across different biological subtypes, and the association between symptoms and brain networks, while evaluating the predictive power of connectivity features for symptoms.

resultsTwo ASD biotypes emerged with distinct connectivity and behavioral characteristics. Biotype1 showed more severe clinical symptoms and a brain network organization marked by stronger global integration but weaker local segregation, reflected in higher global efficiency and reduced clustering, alongside diminished connectivity in core networks such as the default mode network (DMN). Connectivity outside the frontoparietal network (FPN) was positively associated with communication impairments. Biotype2 showed the opposite pattern—greater local differentiation and reduced global integration—with frontoparietal and FPN–DMN connectivity linked to communication, social, and overall symptom severity. Predictive analyses indicated that different connectivity features were most informative for each biotype.

conclusionThis study preliminarily suggests that ISFC may deepen our understanding of ASD heterogeneity and its underlying neurobiological mechanisms.