Scaling Computational Connectomics

Abstract

Connectomics is an emerging interdisciplinary field at the intersection of neuroscience and computer science that seeks to map and analyze the complex wiring diagrams of brains. Studying connectomes offers profound mechanistic insights into how the structural organization of the nervous system gives rise to its function, deepens our understanding of neurodegenerative diseases, and may even reveal how memories are encoded and stored in the brain. However, generating connectome datasets presents substantial technical challenges in image acquisition, alignment, registration, and segmentation. Furthermore, the resulting petascale datasets demand advanced computational techniques to extract meaningful patterns at scale. Because of these challenges, current state-of-the-art connectomics datasets are both time-intensive to produce and limited in spatial extent–typically not exceeding a cubic millimeter–thereby restricting reconstructions to small organisms such as fruit flies or tiny fragments of mammalian brains. To address these limitations, this thesis introduces a suite of novel computational methods designed to overcome key bottlenecks in connectome reconstruction and analysis, thereby accelerating progress in the field. These contributions range from isotropic microscopy volume reconstruction and machine learning models of neuronal morphology to interactive tools for exploring network motifs in connectomes. By enabling faster and more scalable connectome reconstruction, this work paves the way for large-scale comparative analyses across species, disease states, and developmental stages, and ultimately could advance our understanding of how learning and memory are implemented in the brain.