Bayesian Inference Framework for the Analysis of Biological Systems Applied to RNA Activation in Human Cells

In this work, we present in-depth applications of Bayesian probability theory to several problems of inference that are commonly encountered in modern biological research. The resulting inference framework is applied the analysis of kinetic models of a synthetic RNA activation system in mammalian cells. The main test system to which we apply our analysis involves the little-understood phenomenon of RNA activation, wherein small double-stranded RNAs that are transfected into cells increase the transcription rate of genes whose promoters are similar in sequence to the transfected RNAs. We design, build, and validate the first example of a synthetic promoter to be targeted by RNA activation. To model our RNA activation system, we begin by considering the problem of determining a cell population's growth parameters from cell count data collected over a time interval. We derive an approximation to the posterior probability density for the parameters that characterize an arbitrary model growth curve. We show that our approximation works well in most realistic parameter regimes, and that it can be implemented with good speed on modern processors. Our inference framework is applied to several test datasets in the simple case of a logistic growth model, and its results are compared to the commonly used least-squares estimation method. We then explore the fitting of arbitrary kinetic models to sets of flow cytometry data, which contain distributional information on cell fluorescence. We develop a software system that returns posterior probability densities for kinetic model parameters, and allows for different models to be compared for the quality of their fits. Our software allows the experimenter to probe the logical connections between the various parameters a model, and to draw informed conclusions as to its correctness in light of the data. We briefly discuss the conversion of the known literature values of some parameters into prior probability densities that may inform our reasoning. Finally, we show how to synthesize the solutions to these separate inference problems into a singe coherent conclusion that correctly weights the importance of each information source, and apply this holistic approach to model our synthetic RNA activation system.Ph.D