UBC Theses and Dissertations
Identifying interaction effects in high-throughput studies of growth Ushey, Kevin
In genomics, a newly emerging way to learn about gene function is through growth curve experiments. In such experiments, different strains of yeast (Saccharomyces cerevisiae) – single mutants having one gene knocked out, double mutants having two knocked out – are grown in microtitre plates, with an automated system capturing the size of cell populations over time. These growth curves can provide information on the function(s) of the associated genes. Of particular interest are interaction effects, where the growth of a double mutant is surprising in light of the growth of normal yeast and its two corresponding single mutants. There is currently a lack of consensus on the best way to analyze growth curve data. For a growth curve experiment, strain fitness must be defined in some way in order to separate and rank strains according to their ability to grow, and it is uncertain which possible definitions of strain fitness have better ability to identify real interaction effects than others. After defining strain fitness, this quantity must be estimated for each strain through either parametric or non-parametric model based approaches, and the approach used can also affect the ability to identify interaction effects. Furthermore, different problems related to the experimental protocol present themselves when attempting to model growth curves, and these need to be accounted for as well. In this thesis, I will explore and compare some commonly used models and definitions of strain fitness when analyzing growth curves, and relate them concretely to the exponential and logistic models upon which they are built. I will compare and contrast multiple methods used when attempting to analyze growth curve experiments, and seek to propose Area Under the Curve as a definition of strain fitness which prompts a derived variables modeling strategy that performs well in ease of implementation, retains flexibility in assessment of a heterogeneous mix of sigmoidal and non-sigmoidal growth curves, and remains able to identify interaction effects.
Item Citations and Data
Attribution-NonCommercial-ShareAlike 3.0 Unported