George Pampoukis

The growing demand for safe, nutritious, and high-quality food that is produced with limited environmental impact and minimal food loss and waste requires preservation technologies that reliably control microbiological hazards without unnecessarily compromising product quality. Conventional thermal processing has a long history of food applications but can negatively affect sensory and nutritional attributes. Mild or non-thermal technologies such as non-thermal plasma (NTP), pulsed electric field (PEF), and high-pressure processing (HPP) are increasingly proposed as alternatives that may combine high levels of microbial safety with improved product quality and, in some cases, lower environmental burdens. At the same time, food business operators and risk assessors need quantitative evidence on decontamination performance across products, microorganisms, and process settings in order to design and validate control measures, perform risk assessments, and compare technologies on a fair basis. This thesis aimed to obtain quantitative, generic insights into the decontamination efficacy of NTP, PEF, and HPP under food-relevant conditions by systematically collecting, structuring, and analyzing existing literature data. Rather than optimizing a single process or product, the work focused on extracting global kinetic parameters and identifying the key processing, matrix, and microbial factors that drive variability in inactivation outcomes. To do so, it combined meta-regression and other statistical approaches with recent data-science methods, including active learning for literature screening, transparent data-wrangling pipelines, and machine-learning algorithms to obtain generic insights. The resulting benchmarks are intended to support process design and validation, quantitative microbiological risk assessment, and technology comparison in a broader sustainability context. Chapter 2 addressed non-thermal plasma decontamination in food-related applications. A literature-derived database of log10D-values was compiled across a wide range of plasma sources, matrices, and target microorganisms. Because the data were highly heterogeneous and reporting practices differed considerably between studies, classical kinetic meta-regression was not feasible at a generic level. Instead, the chapter introduced dissipated power per plasma volume as an integrated descriptor of process intensity for dielectric barrier discharge configurations and used machine-learning methods to rank the relative importance of process, product, and microbial factors. Power density and matrix category emerged as the most influential predictors of decontamination efficacy, followed by microbial genus and pH. The chapter also demonstrated, for the first time in food microbiology, the use of active learning to make literature screening more efficient, reducing the manual workload while maintaining coverage of relevant studies. Chapter 3 responded to the lack of standardization that limited the analyses in Chapter 2 by developing a dedicated metadata schema for NTP decontamination in food-related applications. Building on the existing Plasma-MDS framework from plasma science, the new schema defined a structured set of parameters covering the plasma source, working medium, food or target matrix, microbial characteristics, and diagnostic methods. The schema was implemented in human- and machine-readable format and illustrated with concrete examples how experimental parameters should be reported. By making key experimental details explicit and interoperable, this work supports FAIR (findable, accessible, interoperable, reusable) data practices for NTP and lays the foundation for future meta-analyses and predictive modelling once more consistently reported metadata have accumulated. Chapters 4 and 5 focused on PEF and HPP, respectively, for microbial decontamination in food products. In Chapter 4, a harmonized database of 569 PEF experiments was compiled for seven microbial groups, including E. coli, Listeria spp., Pseudomonas spp., S. aureus, S. cerevisiae, Salmonella spp., and lactic acid bacteria (LAB), across multiple liquid food matrices. Decontamination efficacy was expressed using the decimal-reduction volumetric energy input (DEvol), defined as the energy per litre required for a one-log10-reduction. Meta-regression was used to estimate global and matrix-stratified DEvol values, revealing that matrix composition can change the decimal-reduction energy by up to eight-fold. High-acid fruit juices showed comparatively low DEvol values, whereas high-acid alcoholic products were more protective, demonstrating that acidity alone is insufficient to describe matrix effects. A decision tree was developed to provide estimates of the energy input needed to achieve a 5-log10-reduction for different microorganisms and matrices, and an interactive dashboard was created to allow users to visualize and explore the underlying data in real time. In Chapter 5, high-pressure processing was examined using an extended database of 2692 log10D-values for 15 microbiological hazards treated at temperatures ≤ 45 °C. Using a Bigelow-type model fitted to the available data, global log10D-values were estimated at a reference pressure of 500 MPa, allowing microbiological hazards to be grouped into three resistance categories (sensitive, moderate, and resistant) at this pressure level. The analysis confirmed that matrix properties strongly influence HPP resistance: low water activity (aw ≤ 0.95) tended to protect microorganisms, while acidic or aqueous matrices (pH ≤ 4.5 and aw > 0.95) facilitated inactivation. Multiple linear regression indicated that aw and pH categories were the dominant product-level predictors of inactivation, whereas other variables such as protein and carbohydrate content had comparatively limited effects. Together, these results led to a decision-support framework combining (i) hazard resistance ranking at 500 MPa, (ii) matrix-stratified inactivation estimates by food category, and (iii) multiple linear regression models based on aw and pH, providing flexibility depending on the level of information available to users. Chapter 6 integrated the technology-specific findings into a broader food safety and sustainability perspective by comparing thermal processing, HPP, and PEF for the inactivation of E. coli in high-acid fruit juices. Using meta-regression models developed in this thesis, the chapter derived efficacy benchmarks (log10-reductions) and their variability for each technology under various industrially relevant processing conditions reported in the literature. These efficacy estimates were linked to energy use and, where available, greenhouse gas emission data from the literature. The analysis illustrated that conclusions about “more sustainable” technologies depend critically on matching decontamination performance across processes and on considering both energy and GHG emissions per log10-reduction rather than per litre alone. It also highlighted that electricity-driven mild technologies may become relatively more favorable as electricity production relies more and more on renewable sources. However, although thermal processes with heat recovery compromise fresh-like properties, they can remain highly efficient from both an energy and a decontamination performance perspective. Finally, the chapter outlined how the meta-regression framework used here could be coupled with life-cycle assessment and extended to other technologies and product-pathogen combinations. Taken together, this thesis shows that carefully curated databases, transparent data processing pipelines, and quantitative synthesis methods can extract generic, practically useful information from heterogeneous literature on mild food processing technologies. For NTP, it clarifies which factors most strongly influence decontamination and introduces a metadata infrastructure to enable more robust comparisons in the future. For PEF and HPP, it delivers global kinetic parameters, hazard and matrix resistance rankings, and simple tools such as decision trees and dashboards that can support process design, validation, and risk assessment. By bringing these pieces together in a comparative framework that explicitly considers both microbiological performance and sustainability factors, the work links data-driven modelling to practical decision-making on process choice. In doing so, it contributes to the evidence base needed to deploy mild preservation technologies in a way that is scientifically justified, transparent, and aligned with broader ambitions for safe, lower-waste, and more sustainable food systems.