Abstract
Understanding which volatile compounds discriminate between blue cheeses is important to maximise product quality, facilitate innovation, and determine authenticity. However, identification of discriminant volatiles between blue cheeses has been hindered by methodological limitations and the presence of non-linear volatile behaviours during blue cheese ripening. To address these challenges, an end-to-end analytical workflow was optimized and a new non-linear data analysis approach using self-organizing maps and entropy-based feature selection (SOM-EFS) was adapted from the field of machine learning. To validate the SOM-EFS methodology, its predictive performance was compared throughout the thesis with an established linear chemometric approach. Three untargeted volatile fingerprinting experiments, each using solid phase microextraction, gas chromatography, and mass spectrometry, were conducted to identify discriminant volatiles i) between blue cheese varieties; ii) as a function of ripening time in two varieties of blue cheese; and iii) between cheeses produced using different Penicillium roqueforti strains, curd compositions, and ripening times/temperatures. The non-linear SOM-EFS technique effectively discriminated between samples and yielded better predictive performance than the linear models. However, there were interpretive synergies by utilizing the two approaches in tandem. Across all experiments, it was determined that alcohols especially, but also esters, ketones, and sometimes even hydrocarbons, served as effective discriminant compounds in blue cheeses. As a result of the first experiment, 1-nonene and 2,6-dimethylpyridine were reported for the first time in blue cheeses. The second time-series experiment confirmed, and effectively modelled for the first time, the non-linear volatile behaviours in blue cheeses during ripening. The second study also identified two distinct patterns of methyl ketone and secondary alcohol generation and interconversion during blue cheese ripening. The third experiment demonstrated that differentiation of blue cheese volatile fingerprints is driven mostly by selection of the Penicillium roqueforti strain, followed by ripening time, then curd composition, while temperature was found to have little effect on the development of volatiles. This thesis has furthered the understanding of which volatiles differentiate blue cheeses both by variety and as a function of Penicillium roqueforti strain, curd compositions, ripening temperatures, and ripening times. It has also contributed a new non-linear data analysis tool, SOM-EFS, that can be applied more generally in the fields of food and flavour chemistry.