Geometric Representation Learning for Insights Into Cellular and Molecular Behavior
Date of Award
Spring 2024
Document Type
Dissertation
Degree Name
Doctor of Philosophy (PhD)
Department
Computational Biology and Bioinformatics
First Advisor
Krishnaswamy, Smita
Abstract
Large-scale genomics technologies have revolutionized our ability to understand key biological processes, including at the resolution of an individual cell. However, such single-cell technologies present several challenges, including complex and nonlinear interdependencies between features, extremely sparse and noisy measurements, and high-dimensionality. Therefore, sophisticated computational techniques are needed to help us better capture complex and biologically relevant patterns. In this thesis, I demonstrate the utility of geometric representation learning, especially manifold learning, for constructing meaningful representations that capture the underlying structure of single-cell data. First, I describe Gene Signal Pattern Analysis, an approach that learns embeddings of genes, or features, from single-cell RNA sequencing data to better represent local and global gene-gene relationships. Next, I present Directed Scattering Autoencoder, which describes a directed version of geometric scattering for embedding directed knowledge graphs, and show its utility in the context of cellular signaling. Third, I present Archetypal Analysis Network (AAnet) for capturing the continuous variation of the cellular state space at single-cell resolution with archetypal analysis. The contributions of these approaches are demonstrated with synthetic, public, and newly-generated datasets, and we showcase myriad downstream analyses in close collaboration with colleagues across the globe. Such interdisciplinary collaboration played a large role in my PhD, and I thus present a final chapter to highlight the ability of novel computational approaches for investigating cellular and molecular behavior in diverse settings.
Recommended Citation
Venkat, Aarthi, "Geometric Representation Learning for Insights Into Cellular and Molecular Behavior" (2024). Yale Graduate School of Arts and Sciences Dissertations. 1374.
https://elischolar.library.yale.edu/gsas_dissertations/1374
