Regression of Multicomponent Sticking Probabilities Using a Genetic Algorithm

Abstract

A genetic algorithm (GA) was developed for the purpose of regressing composition-dependent aggregation kernels from time series of experimentally measured component or size distributions. The GA evolves initially random populations of kernel models in accordance with the principles of microevolution. To test the robustness of the GA, functionally diverse kernelsincluding one describing the shear-mediated aggregation of blood cellswere constructed. The stochastic time evolution of their corresponding aggregation processes were then simulated under physiological conditions via Monte Carlo. The GA successfully regressed the kernels underlying these “gold standard” datasetswhere we employ the term in the sense of a “trusted reference”from these simulation results, reproducing the multicomponent kernels to a maximum relative deviation of less than 9% over their entire composition ranges. Finally, ramifications of these cases pertinent to experimental design are considered, including the effects of extreme initial population ratios for multicomponent aggregation experiments with extreme population ratios.