Neurobiological and Neuropsychological Deep Endophenotypes of Behavioral Response Inhibition: RDoC Empirical and Mathematical AI Feature Engineering
Loading...
Authors
Wehrli, John
Issue Date
2025-09
Type
Dissertation
Language
en
Keywords
Behavioral Neuroscience , Artificial Intelligence , Psychology , Endophenotypes , impulsiveness , impulsivity , Research Domain Criteria , RDoC , digital twins , schizophrenia , autism , ADHD , borderline personality disorder , response inhibition , computational psychiatry , psychiatric genetics , behavioral genetics , Business, Engineering, Science, & Technological Innovation , Healthcare Innovation & Delivery
Alternative Title
Abstract
Impulsivity and behavioral response inhibition represent core transdiagnostic constructs in behavioral neuroscience and translational psychiatry, central to ADHD, autism, borderline personality disorder, and schizophrenia. Despite decades of research, the literature fails to adequately characterize cognitive control dimensionality, limiting differential diagnosis and precision psychiatry. This quantitative Research Domain Criteria (RDoC) study discriminated and dimensionalized neurobiological and neuropsychological endophenotypes through deep phenotyping, guided by Barkley's inhibitory deficit theory, to enable computational neuroscience applications. 480 adults (of 6,603 participants) completed the Barratt Impulsiveness Scale (BIS-11), Wisconsin Card Sorting Test (WCST), Go/No-Go Task (GNG), and Stop Signal Task (SST). Participants were dichotomized into high and low trait motor impulsivity (MI) groups. MANOVA examined group differences across nine cognitive control variables. Cook's influential outlier analysis enhanced statistical power, while cross-model convergent analysis identified optimal deep endophenotype combinations for artificial intelligence (AI) machine learning feature engineering. Results demonstrated significant multivariate WCST effects, p = .009, partial η² = .024, with marginal significance after influential outlier removal, p = .063, partial η² = .018; significant multivariate GNG effects, p = .008, partial η² = .028, with practically significant SST effects, p = .019, partial η² = .024. Cross-model convergent analysis identified three optimal decomposed endophenotype candidates (WCST non-perseverative errors [NPE], GNG commission errors [CE], SST stop signal reaction time [SSRT]) for discriminating MI groups, p < .001, partial η² = .040. A reaction time threshold discovery revealed high MI individuals achieve optimal GNG CE minimization significantly faster (558ms/high-MI vs. 625ms/low-MI), challenging conventional intervention approaches by suggesting strategies based on speed-accuracy optimization. WCST findings also discriminated motor and choice impulsivity subtypes through preserved perseverative responses (PR), suggesting intact caudate nucleus function while revealing MI-specific prefrontal dysfunction. This computational and behavioral neuroscience study successfully validated deep decomposed endophenotype candidates, resolving measurement problems in impulsivity research, providing empirical support for RDoC latent variable approaches. Findings enable neurocomputational modeling applications including machine learning, deep learning, and digital twins for behavioral phenotype assessment in computational psychiatry. Future research should validate these endophenotype (neurophenotype) candidates across independent samples, expand to genetic and neuroimaging studies, and develop neurocomputational AI models for personalized psychiatric interventions.