Project Abstract
The effect of analytic flexibility on brain-behavior relationships and predictive models of adolescent
socioemotional processing is not well understood. The Maturational Imbalance (or Dual System) Model often
lacks reliability and generalizability. Existing work has predominately focused on single task-designs and small
samples (median < 50) concentrating on brain-behavior associations using disparate operationalizations of
reward and affective processing. The proposed research will integrate three developmental functional magnetic
resonance imaging (fMRI) samples (N ~ 105; N ~ 180; N ~ 7,000), with analogous reward and affective
paradigms, to investigate key issues related to reproducibility and generalizability: (a) the influence of analytic
flexibility on brain-behavior associations and convergence and predictive validity in contrasts within/between task
domains; and (b) uncovering task-based fMRI (t-fMRI) brain features (latent neural characteristics) that can serve
as the basis for robust brain-behavior prediction models across multiple samples. It is hypothesized that t-fMRI
contrasts can be separated across a multidimensional plane of attention and valence, which elicits neural
responses leading to approach or avoidance. However, how researchers operationalize positive and negative
valence in t-fMRI often varies, and this variability in the decision-making process may influence the underlying
neural effects. Aim 1a will examine how brain-behavior associations in a given task change based on analytic
decisions relating to fitting general linear models (GLM), contrasts and neural regions. Then, Aim 1b will consider
whether changes in brain-behavior associations (as a functional of analytic flexibility) are reflected in changes in
construct validity of approach and avoidance within- and between-task domains, such as reward and affective
processing. Conversely, traditional univariate GLM approaches show mounting issues in test-retest reliability
and express associations that may not support generalizable prediction of behavioral phenotypes. However, the
neurodevelopmental literature has proposed that multivariate analyses that leverage dimensionality reduction
and machine learning can provide informative brain-behavior prediction models. To test this hypothesis, in Aim
2, dimensionality reduction will be used in a large adolescent t-fMRI sample to generate brain-behavior prediction
models and compared across a reward and affective task to consider the influence of constructs. Aim 3 will focus
on the dissemination of code and fMRI statistical maps. The fellowship will support the applicant's growth in
becoming an independent researcher and leader in the neurodevelopmental neuroscience by providing training
in: combining t-fMRI datasets, evaluating the effect of analytic flexibility in fMRI and impact on construct validity,
applying dimensionality reduction in neurodevelopmental samples to produce brain-behavior prediction models.
This training will support the applicant's long-term goals of understanding of neural mechanisms in adolescent
substance use and improving our understanding of traditional and non-traditional measurement models.