UCSF

Tissues comprise a multiplicity of specialized cell types that must coordinate state changes in order to function collectively. These state changes are orchestrated by coordinated direct cellular interactions and indirect responses to microenvironmental and systemic cues. Consequently, chronic perturbations to this collective behavior can result in disease states that are difficult to reprogram such as autoimmunity and cancer. As such, studying the self-reinforced dynamics of tissue function can benefit from a systems biology approach where the aim is to understand how individual components of biological systems interact to give rise to emergent properties. The recent growth in the availability of single-cell resolution genomics platforms has further expanded biologists’ ability to do this kind of unbiased inquiry. However, despite the increasing ease of generating these high-dimensional datasets, analyzing these data still presents significant computational challenges because of their noise and sparsity, which are further exacerbated on the level of individual cells and genes. As such, there is a need to develop computational methods that enable scientists to extract systems-level biological insight from noisy high dimensional data. This dissertation introduces DECIPHER, a machine learning framework tailored for network inference in systems biology, with a focus on applications to single-cell RNA sequencing data. Chapter 2 details the DECIPHER algorithm and its implementation for the R computing environment, deciphR, that is designed to reconstruct cell state networks from high-dimensional molecular profiles. Chapter 3 applies DECIPHER to unveil cell-cell interaction networks in the human breast, elucidating how state changes on the cell-level propagate throughout tissue in response to hormonal fluctuations. Chapter 4 extends DECIPHER's application to investigate peripheral immune dysregulation in a rare pediatric autoimmune disease, revealing underlying immune imbalances that persist even in disease remission and potential therapeutic targets. Overall, this dissertation presents a generalizable approach to network inference for systems biology and demonstrates its utility in multiple biological contexts for unravelling cellular coordination in tissue homeostasis and disease.