(313f) An In-Silico Evolution Algorithm for Obtaining Parameter Values In Stochastic Kinetic Models
Stochastic biological models can describe a range of phenomena that are to a certain extend random, such as emergence of cancer cells, apoptosis or emergence of drug resistance. A concern with these models is the large number of model parameters that may not be easily determined by traditional least square techniques used for deterministic models.
In the present work, several stochastic models of the budding yeast cell cycle are presented and approaches to estimate the model parameters are discussed. We have found that in-silico evolution is a potentially useful tool for obtaining model parameter values that give biologically reasonable results. Using this algorithm, growth of a cell population is simulated such that the stochastic parameters are allowed to change in each division while the population size is kept constant by continuous removal of cells based on some selection criteria. The selection criteria can be as simple as removal of cells at random, in which case the algorithm selects for the faster growing phenotype.
The simulation results show that the average cell mass converges rapidly to a steady value while many of the stochastic parameters continue to fluctuate. Presumably, large fluctuations indicate that the value of a parameter is not important in determining the phenotype while small fluctuations indicate the opposite. The magnitude of the allowable change in parameter values at division affects the execution time of the algorithm and convergence to a steady value and allowing the changes to be too great results in excessive cell death.