Browsing by Subject "Bayesian data analysis"

Asymmetrical flow field-flow fractionation (AF4) is a separation and characterization technique for macromolecules and particles, which has been gaining popularity in a multitude of scientific and industrial applications. AF4 is considered a challenging experimental technique to optimize and relatively few tools exist for this purpose. One of the main aims of the work was to provide practitioners of AF4 techniques with software tools, which bridge the gap between domain knowledge and AF4 theory for a more fluid experimental design workflow. This is made possible by a feature of AF4, which makes it stand out from related separation methods by enabling theory-driven prediction of sample behavior over the course of the experiment. In the first part of the computer experiments carried out, an algorithm based on probability theory was developed for predicting the ideal separation of samples based on readily obtainable sample properties. Among the obtained results is a predicted fractogram, which is the end product of an AF4 experiment run. The ability to predict separation of samples finds use in AF4 method development as well as other applications relevant to experimental work. The algorithmic models were constructed to describe real life systems for which experimental data was available and against which performance could be tested. The real world systems modeled included two AF4 instrument channels with different geometries and both natural and synthetic polymer samples. Prediction by the algorithm was compared to previously published experimental data from other authors, after configuring the algorithm to the corresponding experimental setups. The results suggest that the algorithm can relatively closely approximate predictions made by the underlying ideal AF4 theory. For a disperse polymer sample in a separation program for which no simple theoretical result was available, the algorithm's predictions gave promising results for approximating the shape of fractogram curve. In the second part of the computer experiments, a theory based model was fitted to experimental data and performing statistical inference was explored as a technique. Bayesian data analysis was used therein to complete a routine task in AF4 operation and subsequent data processing. The data analysis provided an estimate congruent with theory and external estimates given for the same data by other researchers. Looking forward, possible enhancements to the presented models and their applicability more widely to AF4 work as well as possible developments of computational models in the field are discussed.