Project Summary Autism spectrum disorder (ASD) is a developmental disorder characterized by impairment of social interaction and communication, as well as repetitive behaviors, with severity ranging from mild to significantly disabling. The prevalence in the United States is rising (currently about 1 in 54 children) and the associated costs are enormous. In our most recent previous efforts on this project, we have advanced methods for the analysis of functional magnetic resonance imaging (fMRI), using both task-based and resting-state data, for classification, facilitating identification of ASD biomarkers, and using these biomarkers for personalized outcome prediction for Pivotal Response Treatment (PRT), an evidence-based form of behavioral therapy for ASD. We have made great strides with our most recent work focusing on deep learning techniques for extracting biomarkers and predicting outcome using novel strategies focusing on temporal characteristics with Long Short Term Memory (LSTM) networks as well as spatial characteristics using Graph Neural Networks (GNNs). Improved use and characterization of the dynamic changes in connectivity appear crucial for advancing performance based both on our work and the literature. Thus, in our proposed work, we intend to develop a richer, integrated model that can more fully exploit the complete spatiotemporal characteristics of the data and its inherent dynamics. In addition, a key issue in deep learning strategies is access to large datasets which remains a challenge, especially given that ASD is a spectrum with a range of characteristics, severity and comorbidities and that fMRI is a powerful tool that encompasses many task-based and resting state acquisition paradigms. Comorbidities are a particular challenge and opportunity in that they have the potential to increase our understanding of the heterogeneous manifestations of ASD. We propose to further develop and expand the power and impact of our methodology by broadening our subject base to include ASD comorbidities (e.g. anxiety, ADHD, depression) and multiple treatment strategies. We will advance our technology by creating a more integrated spatiotemporal analysis combining our LSTM and GNN approaches. We will use domain adaptation methods to properly exploit multiple fMRI paradigms in conjunction with secure federated learning strategies to facilitate multi-institutional data usage with privacy. These innovative approaches will allow us to improve the practicability of predicting quantitative treatment outcomes in ASD and measuring associated neuroimaging biomarkers.