Improving ms-sensor technologies for food quality assessment

Resumen

Food industry is demanding for fast screening methodologies in order to guaranty food quality and safety. These methodologies should allow high throughput with sufficient accuracy and reproducibility. In this context, MS-Sensor is a challenging approach since it allows simultaneous determination of compounds in food matrices and complex mixtures with a high sample throughput. The basic working principle of MS-Sensor systems is based on the introduction of volatile components extracted from the headspace of a sample into the ionization chamber of a mass spectrometer. The mass spectra resulting from the ionization and fragmentation of this extract constitute a very complex ionization pattern that can be seen as a 'fingerprint' which is characteristic of the matrix being analyzed. These ionisation patterns are then processed by pattern recognition engines to perform tasks such as classification, recognition and, to a limited extent, quantification.This thesis is devoted to study the possibilities and capabilities of MS-Sensor approach to its application in several food quality related problems such as the determination rancidity levels in crisps; the detection of fungal spoilage in bakery products, the monitoring of sardines freshness under cold storage; the classification of virgin olive oils according its organoleptic properties and the discrimination of two Iberian ham qualities according pig's feeding. In each one of these applications it has been demonstrated the feasibility of using a MS-Sensor to solve the food quality problem under study. It has been widely demonstrated that the MS-Sensor profile can be considered as a useful fingerprint technique for characterization of the targeted quality property and, as in certain cases, even for quantification of several parameters correlated with this problem.Despite that feasibility of MS-Sensor has been widely demonstrated for the applications under study, this approach stills suffer from some weakness or drawbacks that may influence performance of MS-Sensor. Main drawbacks are the inherent high dimensionality of data response matrices and the low selectivity of m/z fragments pseudosensors used as variables in these matrices.These two drawbacks could be responsible for the lack of reproducibility showed by MS-Sensor systems in certain applications mentioned above. In order to success in the development of such applications it was necessary to figure out different strategies for overcoming this high dimensionality and the low selectivity. In order to handle the low selectivity of m/z fragments several new methodologies based on the use of multi-way algorithms has been implemented for the first time in the framework of this thesis. Besides, new variable selection algorithms has been developed and implemented in order to avoid high dimensionality modelling. It has been demonstrated that use of the developed algorithms leads to a simpler and more parsimonious models and consequently to a better performance and more reproducible results.In addition, several issues related to use of MS-Sensor in food analysis has been studied: the use of different headspace sampling techniques; the comparison of MS-Sensor systems performance against classical MOS based 'electronic noses'; the application of new algorithms for pre-processing MS-Sensor signals; the correlation of MS-Sensor response and the well-established methods to assess the quality property under study, etc.MS-Sensor is a powerful device set-up, capable of producing large amounts of highly selective information. Optimal use of this device implies both, a correct use of analytical techniques (sample handling and instrumental) and a rational use of subsequent data analysis. That can be only attained if analytical people in charge of experimental set-up work side by side with data analysis and software developers. This thesis aims to bring nearer this close collaboration.