Date of Award
Doctor of Philosophy
Steven M. Abel
Tian Hong, Stephen J. Paddison, Cong Trinh
The use of computer simulations in biology is often limited due to the lack of experimentally measured parameters. In these scenarios, parameter exploration can be used to probe biological systems and refine understanding of biological mechanisms. For systems with few unknown parameters, parameter sweeps that concurrently vary all unknown parameters are tractable. In complex systems with many unknown parameters, supervised machine learning algorithms can be used to discover parameters leading to targeted system responses. In this thesis, we study three biological problems in which we use parameter exploration methods to gain mechanistic insights. We first explore the role of altered metabolism in cancer cells that reside in heterogeneous tumor microenvironments. We use a multiscale, hybrid cellular automaton model to evaluate tumor progression while varying malignant cell traits using a systematic parameter sweep. The results reveal distinct growth regimes associated with varied malignant cell traits. We then study kinetic mechanisms governing fixed-topology signal transduction networks and use evolutionary algorithms to discover kinetic parameters that produce specified network responses. We analyze the growth-response network in Arabidopsis with this supervised machine learning approach. This allows us to identify constraints on kinetic parameters that govern the observed responses. The evolved parameters are used to calculate the responses of individual network components, which are used to generate hypotheses that can be tested in vivo to help determine the network topology. We finally apply a similar approach to redesign signal transduction networks. We demonstrate that the T cell receptor network and an oscillator network show remarkable flexibility in generating altered responses to input, and we further use a nonlinear clustering method to identify design criteria for the underlying kinetic parameters. For each project, observations produced from in silico simulations lead to the formation of hypotheses that are experimentally testable.
Prescott, Aaron Matthew, "Characterizing Signal Transduction Networks and Biological Responses Using Computer Simulations and Machine Learning. " PhD diss., University of Tennessee, 2018.