Autism spectrum disorder (ASD) is a neurodevelopmental condition that heterogeneously impacts core domains of human function, such as communication, socialization, and cognition. Underlying these symptoms are complex genetic changes in a diverse array of genes and pathways. Developing a mechanistic understanding of ASD remains a major challenge that, left unanswered, will continue to hinder progress in the clinical management of this increasingly prevalent disorder. New approaches are needed to improve our understanding of the neurobiology of ASD, as well as current clinical paradigms.

In this dissertation, I discuss how techniques from the quantitative fields of machine learning and dynamical systems theory may be applied to decipher the cellular and molecular complexity of ASD. In Chapter 2, I address how whole genome sequencing data can be analyzed through machine learning to predict ASD phenotypes. The machine learning analysis describes novel methodology to vectorize genomics data and build interpretable classification models that relate to well-supported findings in the literature. In Chapter 3, I show that neuronal activity can be modeled as a nonlinear dynamical system to yield novel measures of neuronal state and dysfunction. I provide evidence for the minimum embedding dimension (MED) as a marker of diminished dynamical complexity of electrical activity recorded from ASD patient-derived neurons. I also probe the clinical and gene expression correlates of MED, which overlap with known ASD risk genes and pathways related to neurodevelopment in cortical and deep brain structures.

The approaches outlined in this dissertation seek to transfer tools from different quantitative fields to identify convergent cellular and molecular findings that characterize ASD. Genome classification and dynamical analysis improve on existing methods for the identification of disease signatures and reveal a set of common biological pathways and brain regions in ASD.