Beyond the average: computational approaches for analyzing and modeling cellular variability
This thesis presents a set of novel statistical methods that advance the analysis of phenotypic variability, the phenomenon where genetically identical cells display different behaviors even when sharing the same environment, from both exploratory and mechanistic perspectives.
The first contribution is a method called DNGP, which makes it possible to statistically pinpoint when two groups of cells, such as wild-type and mutant strains, behave differently over time. This is followed by the development of two mechanistic modeling techniques, GMGTS and EGM. These methods significantly accelerate the process of identifying biological "rules" or parameters by fitting models to the rate of change in cell data, which is much faster than traditional simulation methods where the cell measurements themselves are matched. GMGTS and EGM are especially effective for complex systems where certain biological components are difficult to measure directly.
The proposed statistical frameworks of this thesis form a practical toolkit for scientists to study cellular individuality and helps bridge the gap between noisy single-cell data and clear biological insights. Overall, these methods provide efficient alternatives to existing techniques, facilitating a more rigorous understanding of how natural biological noise influences cellular functions like growth and division.