Recent advances in connectomics have produced large-scale datasets aligning brain connectivity with genetic expression, and these resources are poised to grow dramatically with ongoing technological and institutional investment. In response, researchers have developed computational models linking gene expression to brain wiring, often using regression techniques that identify genes implicated in neural development. While these models offer valuable biological insights, they are inherently constrained: they explain connectivity only as a function of known gene expression patterns, limiting the kinds of hypotheses that can be formulated about how connectivity arises. This dissertation introduces a generative modeling framework that reverses the traditional direction of explanation. Instead of predicting connectivity from genetics, it constructs connectivity from first principles, using unsupervised latent variables that are fit to explain observed wiring patterns. The model reveals strong statistical associations between learned latent representations and genetic expression, especially under sparse and low-complexity assumptions, suggesting that gene expression encodes the structural motifs of connectivity. Building on this foundation, a developmental model is introduced based on the theory of morphogenesis. This model generates latent variables through a sequential process governed by local signaling rules, capturing the constraints of biological development. It is found that the developmental model explains even more of the observed genetic expression than the static model, providing evidence that gene expression contains disambiguating signals that support pattern formation, cell identity, and positional inference. Together, the results support the theory that brain wiring arises from a lowdimensional, structured genetic program optimized under biological constraints. These models not only advance understanding of neural development, but also open new directions for engineering applications in AI, regenerative medicine, and biologically inspired design.

PDF

PhD Thesis, Neuroscience, The University of Texas at Austin, Austin, TX, 2025.

@phdthesis{warner:phd25, title={Developmental Generative Models of Brain Connectivity}, author={Jamieson Warner}, month={ }, school={Neuroscience, The University of Texas at Austin}, address={Austin, TX}, url="http://nn.cs.utexas.edu/?warner:phd25", year={2025} }