Identification of outcome-based sub-populations using deep phenotyping and precision functional mapping across ADHD and ASD

点击翻译按钮获取中文摘要

Project Summary Two of the earliest onset, most common, and costly neurodevelopmental disorders in child psychiatry are Attention Deficit Hyperactivity Disorder (ADHD) and Autism spectrum disorders (ASD). The clinical heterogeneity and the imprecise nature of their nosological distinctions represents a fundamentally confounding factor limiting a better understanding of their etiology, prevention, and treatment. In short, simple design assumptions regarding `homogeneity in samples' in typical and atypical populations may explain the frequently very small effect sizes in psychopathology research. Clinically, these same assumptions may account for why treatments often have weak or unpredictable effects. Recent developments in the computational sciences, have enabled the implementation of models sufficiently complex to address the aforementioned situation regarding subpopulations; however, very few tie the outputs to the specific outcome or questions being asked by the investigator. Under the parent grant, we developed and published a novel hybrid supervised/unsupervised machine learning method to characterize biologically relevant heterogeneity in ADHD and/or ASD – the Functional Random Forest (FRF). The hybrid FRF combines machine learning and graph theoretic analyses in order to identify population subtypes related to the clinically most important outcomes (in the case, of this proposal, negative valence symptoms) trans- diagnostically (ASD, ADHD, TD). Despite developing the FRF, subtyping results using functional MRI (fMRI) signals have lagged behind the subtyping of behavioral profiles. In addition, they have yet to become sufficiently sensitive and specific, for rapid translation into clinical practice. Fortunately, parallel advances in functional neuroimaging, allow for precision functional mapping of individuals, and can be synergistically combined with the FRF to greatly boost our ability to subtype and characterize individual patients from fMRI data. Here we combine the FRF with precision mapping to reveal common variants and individual specificity in global brain organization. The proposed individual-specific precision mapping moves beyond group averaging approaches, which are obscuring important inter-individual differences related to distinct pathophysiologies underlying negative valence across diagnoses (ADHD, ASD, TD). Thus, the current proposal aims to apply FRF algorithms to trans-diagnostic (TD, ASD, ADHD) behavioral and precision functional mapping RSFC data to identify distinct sub-populations across ASD, ADHD, and TD that relate to negative valence symptom dimensions.