Using machine learning to link neural structure, genetics, and cognition in a neurodiverse population

The purpose of this research is to gain a mechanistic understanding of why neural structure and cognitive outcomes

differ between people, creating the neurodiversity we see in the population today. This will be done by identifying genes

implicated in phenotypic brain structures in the CALM cohort, a transdiagnostic longitudinal study with fMRI, genetic, and cognitive data of 1000 adolescents. The research first aims to produce polygenic scores for each participant, which are the probabilities that a participant will have a brain phenotype given their genetic makeup. Using these polygenetic

profiles, the research aims to find commonalities between participants using cluster analysis, a machine learning technique used to explore data heterogeneity. This produces genetically common groups relating to brain phenotypes including white matter density, surface area, and cortical thickness. Further, the research aims to produce equations

using cost-value wiring parameters based on generative unsupervised machine learning models that outlines the neural structure of each CALM participant's fMRI data. Using the genetic profiles, it is possible to look at wiring parameter variability within each cluster group and identify the genetic basis of neural structures, as modelled using simple

equations.