Development of a SOftware tool for Prediction<br/>of ready-to-eat food product sHelf life, quality and safetY

Informacje na temat projektu

SOPHY

Identyfikator umowy o grant: 289053

Projekt został zamknięty

Data rozpoczęcia 1 Lutego 2012

Data zakończenia 31 Stycznia 2015

Finansowanie w ramach

• Specific Programme "Cooperation": Food, Agriculture and Biotechnology

Koszt całkowity € 3 806 364,86

Wkład UE € 2 879 156,00

Koordynowany przez

VEREIN ZUR FORDERUNG DES TECHNOLOGIETRANSFERS AN DER HOCHSCHULE BREMERHAVEN EV
Germany

Final Report Summary - SOPHY (Development of a SOftware tool for Prediction<br/>of ready-to-eat food product sHelf life, quality and safetY)

Executive Summary:
Convenience and ready-to-eat foods are gaining every year more and more market share– not only in Europe but also on global scale. The trend towards convenience and healthy nutrition has driven sales of pre-packed leafy salads during the last years. New product developments and the broadening of the range of salad varieties have supported this growth. The need to assure safety, counterbalanced with the demand to supply high-quality products at low cost, affords food producers a very small margin of error. Whilst coping with consumer demands for innovative, healthy and low priced products, food producers must balance this with the need to ensure food safety and an appropriate shelf life for their products. This puts the food industry, of which are 99.1% SMEs, under enormous pressure to reduce costs by optimising processes and products whilst maintaining food quality and safety. This is a particular challenge for fresh or minimally processed food like salads where the risk of pathogenic contamination is inherent in the products nature and numerous outbreaks were linked to these product groups.
Based on the knowledge and expertise of project participants and existing scientific results, the project developed a decision making tool in terms of a web-based software for food producers.
This project was particularly aimed at food producing SMEs: They are seldom equipped with a full laboratory which allows evaluation of product shelf life and safety and therefore rely on external laboratories, their own experience or scientific literature. Through adaptation of process steps and product characteristics in the SOPHY software, users can combine various product characteristics and processing parameters and immediately see the anticipated effect on food safety (risk of pathogen survival or possibility of pathogen growth) and shelf life (due to proliferation of spoilage bacteria). Thus product quality could be improved without compromising safety. Furthermore, the user-friendly SOPHY software could be a valuable tool to support food producers during risk analysis and assessment promoting science based food safety management systems.
The SOPHY project considered concepts of previously developed shelf life and pathogen growth predictors and integrated available data sets. However, the SOPHY prediction software was built on a modular approach allowing combination of various product formulations, processing and post-processing steps. The SOPHY software is therefore able to cover the entire supply chain and overcomes the disadvantage of existing predictors, which are usually considering only one process step. The tool can be used both by salad producers and producers of single components and can be easily extended to integrate further ingredients and adapted to various RTE food products.
The SOPHY software considers food safety (in form of pathogen growth), shelf life (change of spoilage parameters) and impact of processing, product formulation and packaging on quality (organoleptic parameters, nutrients, etc.). The combination of modelling all three aspects is unique and offers food producers new possibilities for product innovation.
The software is supported by information sheets giving background information about product formulation, different novel food processing techniques, as well as hygiene and food safety management. This collection of information combined with predictive quality, safety and shelf life models provides food producers with a useful tool to support product innovation and improvement.
Thus, The SOPHY project aimed to develop and validate a new generation of combined predictive and probabilistic models to be integrated into one single web-based software tool. Furthermore, the system allows users to store their data and results, but at the same time ensures data security by tailored authorisation levels. Thanks to this, SOPHY has a major impact on the establishment and implementation of new food production lines and development of minimal processed, high quality food products. During the project deli salads and fresh cut salads were considered as test case – but the models are transferable to other ready-to-eat food commodities. The project addressed the potential of predictive microbiology and probabilistic models to estimate product safety, shelf life and quality under various conditions. It assists food producers to offer European consumers minimal processed, healthy and high quality products at an appropriate price.

Project Context and Objectives:
Overall aim of the SOPHY project is to develop a software tool for prediction of product safety, quality and shelf life of ready-to-eat products. This tool is mean to support food producers in optimisation of their raw material selection, product formulation and processing steps. In this way, virtual prediction can save time and costs by replacing at least partly time-consuming and expensive laboratory studies.
In order to succeed with this overall mission, the SOPHY project needs to fulfil a number of scientific and technological objectives as defined in Annex I of the Grant Agreement:
1. To develop probabilistic and predictive models, based on existing and newly generated datasets, for:

• Estimation of pathogen growth / no growth
• Estimation of shelf life / food quality optimisation (e.g. colour, texture, viscosity, nutrient retention, organoleptic parameters)

under variable conditions including:
• Processing (examples include washing and use of different types of sanitisers, cutting, modified atmospheres, chilled temperatures including abuse conditions, etc.)
• Product formulation (e.g. changes in pH or aW, additives/preservatives, CIMSCEE stabilized, ‘clean label’ preservatives derived from natural sources)
• Post-production conditions (e.g. temperature during transport and storage, packaging conditions, light, humidity)

dedicated to fresh cut salads and deli-salads

2. To integrate the probabilistic and predictive models into user-friendly software

3. To compile information material for food producers about:

• Food processing technologies (effect on pathogen reduction, change of quality parameters, cost, energy and water consumption, fields of application, etc.)
• Product formulation (effect of change in salt content, pH, use of preservatives etc. on food safety, sensory quality and shelf life)
• Process conditions (effect of process hygiene on food safety, sensory quality and shelf life)
• Food Safety Management (how the kinetic and probabilistic models can be used to support HACCP systems) and process hygiene with reference to relevant legislation


Project Results:
To identifying industry needs and expectations from the software an online survey was created that consisted of a total of thirty-five questions and was available in English and German. The questionnaire was available on the SOPHY project website and distributed via beneficiaries’ networks. The survey was accessed by ninety different companies and eighty-three of the surveys were completed sufficiently to provide meaningful feedback. Furthermore, a total of ten focus groups was conducted groups leading to contact with a total of one hundred and six different industrial representatives. This exercise yielded valuable feedback on the SOPHY software concept whilst raising awareness of the project. In addition, a total of fourteen different food companies were visited on-site and an additional three teleconferences were conducted for feedback on the SOPHY project. A market survey was carried out in order to establish the most popular salads in Europe and their recipe variations. The results of this market survey were discussed and analysed in detail together with salad producing companies of the SOPHY consortium. The registration form for the Advisory Platform was distributed to those companies identified from the online survey, focus groups and factory visits as interested in joining the platform. All Advisory Platform members were mentioned on the project website in order to offering them publicity. The core group members had access to the restricted consortium area of the website where project-related documents can be accessed.
The development of technical sheet templates was mainly influenced by the outcome of the identification of industry needs. The respondents specified comprehensive requirements for information on topics such as microorganisms of concern, preservatives, recommended heat treatments, legislation and HACCP. To meet the needs of industry, the information sheets provide the opportunity to select the product category and the product of concern. A description of the product including its ingredients specifies the product. The information sheets contain comprehensive information about the selected product concerning pathogens, spoilage organism and typical processes. The information about processing is also concerned with storage and transportation. Information about product characteristics such as water activity or pH are captured in the information sheet as well as information about raw material and personnel hygiene.
Three salad products were chosen from each of the food processors within the SOPHY consortium to be included in the studies and assessed to ensure that they represented a wide range of products. The nine products chosen contained salad leaves with and without vegetables; mayonnaise-based deli salads with fish, vegetables, dairy and meat and dessert based deli salads with vegetables.
To define the parameters that have been included in the microbiological and quality modelling, a total of two hundred and eighty-eight independent mathematical equations were retrieved from published reports. For each of these equations, the independent variables were extracted and their interpolation range identified. Then, a systematic evaluation was undertaken to identify the frequency of appearance of each variable, in order to determine the most commonly used for modelling microbial growth and inactivation. In order to store this information, an Excel database was developed to retain these models, including their coefficients.
To specify the overall design of the software, consultation between industry and consortium members has taken place in order to gather feedback in terms of requirements. The output of the first task has been used to guide the software design process. User friendliness has been taken into consideration. The functional behaviour of the software has been documented in terms of activity diagrams, with any anticipated validation and business rules additionally documented. Flexibility has been introduced into the software design by making provision in the design for users to define their own products, data sets, inputs, outputs, models and process chains.
WP2
The data collection related to a broad range of food ingredients and products was performed using a web based database platform (SOPHY Database: http://sophy-wp2.chemeng.ntua.gr/ ). The SOPHY Database is hosted and administered by the NATIONAL TECHNICAL UNIVERSITY OF ATHENS while SOPHY consortium members have full access through their user accounts that were created. The collection of data was based on: (1) Literature (data published in peer reviewed scientific journals or in conference proceedings), (2) Partners’ own unpublished data, (3) Previous research projects and (4) Existing available databases. The objective was to perform a detailed overview of the existing data for the selected food ingredients and products mainly focused on bacterial survival, bacterial growth, spoilage parameters, other quality/sensory parameters under various processing and storage conditions, and shelf life modelling. SOPHY Database collection platform was constructed in seven main fields consist of dropdown lists: product category (characterization and specification), product characteristics, processing (steps and parameters), storage details (packaging and storage temperature range), safety and quality parameters tested (microbiological, physical, chemical and sensory parameters, mathematical modelling: primary and secondary), and shelf-life data (quality parameter used for shelf-life determination, quality parameter characteristics, shelf life values at determined temperatures) and source (existing database, literature, own data, research project)
In order to facilitate an effective data analysis of the collected data, a Search Engine within the Database was developed. The Search Engine was built to give the users the ability of searching within all collected data using search criteria concerning the following fields of interest: product characterization, product category, product specification, processing step, storage temperature range, and packaging. Each search criteria field consists of a dropdown list except for Product Specification field which is actually a free input field. The aim was to give the users the ability of searching within the results, accessing each record data and attached files and retrieving all data concerning the specific records in a zip folder.
Data was collected for a wide range of food ingredients and products with the majority referred as minimally processed, 54% of total records, following by processed ready to eat, 16%, and fresh unprocessed, 15%. 35 and 32% of total data collected was for fruits and vegetables and fresh cut salads, respectively; data referring to deli salads, basically mayonnaise based salads, was rather scarce (5.0% of total data). The fruits and vegetables/fresh cut salads are the following (>0.1% of total entries): apple, artichoke, asparagus, baby spinach, bean sprouts, broccoli, cabbage, cantaloupe, carrot, cauliflower, celery, chicory, cilantro, coleslaw mix, endive, iceberg, irish swedes, leek, lettuce, litchi, lollo rosso, mixed fruit salad, mixed salad, mushroom, orange, pineapple, potato, radicchio, raspberries, red oak leaf lettuce, rocket, romaine, spinach, strawberries, tomato salad, turnip greens, wild rocket; the deli salads most popular are coleslaw dressing, american salad, egg salad, eggplant dip, pasta salad, tarama salad, Russian salad, seafood salad shrimp salad. This data mainly derived by various sources mentioned before; the experimental results produced within SOPHY project were also introduced into the Database (2.8% of total records). Data from SOPHY database was extracted and transferred into a secondary database developed in Microsoft Excel® environment to deal with heterogeneity issues encountered in the SOPHY database.
The systematic analysis of collected data allowed the identification of knowledge gaps that need to be extensively studied through the SOPHY project. Data found in literature was mainly focused on studies per vegetable category and data on mixed salads (consisting of different vegetables) were rather scarce (1.0% of total records). A very interesting aspect of research within SOPHY was decided to be the validation of shelf life data derived from vegetable specific studies in cases of formulating products consisting of several different vegetables. In general most of the collected data on quality and safety parameters of fresh cut salads referred to the effect of storage conditions in these attributes with the majority of them to be developed only under one or two isothermal conditions or more than two but without any mathematically modelling. The lack of systematic mathematical models under variable temperature conditions was recognized so as to be taken into account in the experimental design of the development of quality and safety models. Another aspect regarding mathematical modelling on fresh cut salads was that the majority of published data focuses on mathematical modelling on safety parameters (pathogens such as Listeria monocytogenes, Salmonella, E.Coli). Modelling on quality parameters such as texture softening, colour degradation, enzymatic browning, sensory characteristics was not available. Another issue that emerged through the systematic analysis of collected data was the lack of mathematical models describing the effect of processing conditions (such as washing, cutting) on safety/quality parameters. All the above indicated that even though there are available data on safety, spoilage and quality decay processes (of fresh cut salads), the majority of them is scattered in publications not aimed for development of mathematical models.
According to Database analysis’ results, 34% of the total records included model parameters readily to be used in the SOPHY Software; 66% included unprocessed raw data. All this high value raw data was extracted, mathematically processed (using appropriate kinetic modelling) and gathered in a new Microsoft Excel® file (approximately, 1098 entries for quality and 1030 entries for microbiological indices). Kinetic modelling results were organised in the file as: ingredient/product, microorganism or quality parameter, processing step, model type, independent variable, name and range, output, name and value, equation type and other information regarding differential factor. The newly developed models and the kinetic parameters were then inserted in the SOPHY Database. Users can search for the records with further mathematical modelling in the main screen of the SOPHY Database. The file containing the processed data can also be downloaded.
As a result of the continuous efforts SOPHY Database is now consisted of 75% models for the microbial growth or quality decay, published models (34%) and models produced within the processing (41%); 25% of the published data could not be further processed. Approximately 1339 models were developed for microbial growth/quality loss for various processing/storage conditions to be directly or indirectly used in SOPHY Software.
As mentioned before, data referring to deli salads was rather scarce (95 total records up to now). Almost all records were focused on prevalence data and challenge tests related to pathogens like L. monocytogenes. A few records were also collected dealing with the development of growth/no growth models in mayonnaise based deli salads. From the collected data it was identified that literature is lacking data on the impact of quality parameters such as food structure. Major data gap derived from the analysis was associated with the fact that the literature is lacking of sufficient data on uncertainty aspects governing microbial behaviour, the effect of vegetable type, variety and size of shredding.
The results of this work package served as a basis for design of specific experiments for the data generation, which contributed with data for development and validation of predictive models for spoilage, safety and quality changes.
Due to the high impact of this work package, it was decided to extend the activities involved within the data collection and analysis, up to the end of the project with the goal to keep the Database up to date throughout the entire project duration. This decision was also encouraged and advised by the external experts and the EC project officer during the midterm review process. The work performed within this work package during the last two years of the SOPHY project was more focused in collecting new data while including more parameters on processing, storage and product characteristics. Data referring to complex processing steps were split up into separate unit operations and uploaded to the SOPHY Database as such. The work performed was focused in screening the collected data and excluding any unreliable data from the collection process. All collected data was interpreted and assessed for its validity before it could be further used.
WP3
The main objectives of the data generation were to produce data that could either optimize existing models in the literature or could assist in the development of new kinetic or probabilistic models for the behaviour of spoilage or pathogenic microorganisms and microorganism-independent quality deterioration. The SOPHY consortium identified the following points: (1) Based on literature reports and needs for predictive models challenge tests should include the following types of data:
Independent variables: (i) Time, (ii) Intrinsic factors (pH, aw, preservatives), (iii) Extrinsic factors (temperature, packaging atmosphere, e.g. % composition of CO2, N2, O2)
Dependent variables: (i) Log CFU/g or ml of pathogen, (ii) Log CFU/g or ml of total viable count, (iii) pH, moisture or aw changes of food during storage/processing.
(2) In challenge testing the following principles should be applied: (i) Use of strain-composites:
same strains by different partners, (ii) Activation of strains in TSBYE or TSB with 0.25% glucose or 1% glucose for acid adapted strains, (iii) Inoculation at 6-7 log CFU/g or ml for inactivation studies (e.g. testing efficacy of decontamination interventions, irradiation, HHP), (iv) Inoculation at 1-2 log CFU/g or ml for growth studies, (v) Using common acidulants for pH adjustment, (vi) Using common humectants/drying protocols for aw and moisture control.
Moreover, the quality parameters that were used in order to develop microorganism-independent quality decay models for leafy salads and deli mayonnaise based salads were proposed. Specifically:
• Leafy salads: the most representative quality parameters are texture, color and overall sensory impression and enzymatic reactions involving browning related enzymes.
• Deli salads: the most representative quality parameters are the pH and the viscoelastic properties of emulsions.
Finally, an extensive literature review was conducted and discussions were initiated in order to conclude on a set of independent and dependent variables that could be tested as part of challenge tests and generate data for the optimization of existing models. Emphasis was placed on contributing to the development of SOPs that are relevant to real foods.
The current task aimed: (i) to generate readily available kinetic or growth/no growth data for immediate use in predictive models, as well as (ii) to investigate the role (if any) of “hidden” factors, that affect microbial behavior and lower the performance of existing predictive models.
With regards to the first aim, experimental designs were developed for additional data generation in order to fine-tune and optimize existing models for fresh-cut salads and deli salads or even develop new predictive models for these two food categories.
With regards to the second aim, studies has been performed in the following areas:
(a) Variability in the lag time and generation time of single cells, especially close to the growth boundaries, or after exposure to stresses, (b) Interactions between spoilage flora and pathogens, and (c) Effect of food structure, e.g. solid, semi-solid, liquid, etc. Specifically, the growth variability of low (1-4 cells/sample) or high (1000 cells/sample) populations of Listeria monocytogenes and Salmonella Typhimurium in fresh-cut lettuce and cabbage and their liquid or solidified sterile extracts during storage at 8oC was studied. Inoculum of 1000 cells/sample increased with limited variation (SD <0.5 log CFU/g) on vegetable salads, as opposed to the great variability (<0.7-3.4 log CFU/g) in the growth from 1-4 cells/sample. Total logarithmic increase of 1000 cells/sample of the pathogens on the salads ranged from 1.8 to 2.1 log CFU/g, contrary to 1-4 cells/sample, which exhibited higher increase (2.7-3.4 log CFU/g). The latter suggests that “fail-dangerous” implications may derive from challenge tests with unrealistic high inocula. Different batches of vegetables used for preparation of sterile extracts, introduced high variability in the growth of 1-4 cells/sample, suggesting nutrient-dependent effect on growth of pathogens. Low inoculum of L. monocytogenes did not increase in sterile cabbage extracts, whereas they increased from 1 to 3.5 logs in cabbage salad, probably due to the stimulatory effect of indigenous flora.
In view of the potential needs of the SOPHY Software to provide the users with options for alternative decontamination methods for fresh-cut salads, as well as for mild preservation methods for fresh cut fruit salads, some relevant work was also performed using commercial sanitizers, such as CITROX and mild/natural agents, such as lactic acid, acetic acid and plant extracts. Moreover, the AGRICULTURAL UNIVERSITY OF ATHENS has tested the effectiveness of alginate edible films bearing cinnamon and alcoholic distillates as a means to suppress growth of pathogens, on the surface of apples, pears and banana cubes in modified atmosphere packaging (FOODMICRO 2012).
In an attempt to increase the available options for the SOPHY software users to evaluate the microbial transferability, interactions and adaptive responses, studies in the following topics were performed:
(b) Evaluation of microbial transfer from cutting equipment to vegetables and vice versa, during cutting of fresh-cut vegetables. The objectives of the study were (i) to define the distribution of Escherichia coli O157:H7 and Listeria monocytogenes transfer rates between cutting knives and lettuce leaves and (ii) to model the bacterial transfer from knives to fresh cut salads and vice versa during consecutive cuts of leafy greens. The transfer percentage of E. coli O157:H7 from contaminated lettuce to uncontaminated knives during independent cuts varied from 0.1 to 53.01%, while for L. monocytogenes, the respective percentage ranged from 0.20 to 18.16% on the first day and increased for the 4-days stored leaves up to 79.18%. For both pathogens the distribution was left-skewed. These trends were sufficiently described by the transfer model, showing low RMSE values of 0.799-0.907 and 0.426-0.613 for E. coli O157:H7 and L. monocytogenes, respectively. The model also showed good performance in validation trials with Bf 0.8 -1.5 and Af 1.2-1.5. However, using the electric shredder to cut cabbage, the predicted trends were similar, but the model tended to underestimate the transfer of bacteria, probably associated with the accumulation of contaminated vegetable residues in the shredder. The model also slightly underestimated bacterial transfer during knife- cutting of cabbage and spinach, possibly due to differences in the microstructure of vegetables that might affect bacterial adhesion.
(c) Strain variability and potential fitness of different strains to outcompete other strains in co-culture. The objectives of this study were to investigate the competition among L. monocytogenes or Salmonella enterica strains during growth: (i) on laboratory media and foods and (ii) in selective enrichment. L. monocytogenes strains had similar (p>0.05) growth rates when cultured singly or in mixtures in TSB and TSA. Conversely, growth on ham resulted in cases where a strain did not manage to increase in the presence of another strain. For Salmonella, the presence of competitors in laboratory media, had a significant (p<0.05) effect on the growth kinetics of strains. Depending on which strains were present in a mixture, growth rates of each individual strain could vary in some cases between 0.5 and 1.5 day-1. Strains that were outgrown by others, did not manage to increase more than 7 log CFU/ml or cm2 contrary to the 9 log CFU/ml or cm2 final populations observed in single cultures indicating that maximum populations were also affected by inter-strain competition. During enrichment there were not significant differences among populations of different strains in enrichment (liquid) co-culture. However after streaking, L. monocytogenes 4b serotype outcompeted (80-100% of total colonies) the 1/2a serotype, regardless of the food matrix inoculated before enrichment. In BPW, the dominance (80-100%) of Reading and Putten serovars was reversed in the second enrichment in RVS (0-20% of total colonies).
(d) Effect of structure and habituation under mild acid or osmotic stress on the adaptive responses of Listeria monocytogenes.
Mayonnaise and salad dressing bases are oil-in-water emulsions with potential different rheological properties. In addition, these food products contain organic acids and preservatives to ensure their quality and safety. Although the diffusion of these organic acids is considered uniform, the final pH of a mixed deli-salad (e.g. potato salad, chicken salad, Russian salad, farmers’ salad etc.) may be gradually affected by the buffering capacity of the ingredients. Therefore it is essential to monitor the pH changes in such a complex ecosystem, as it may affect the growth/no growth boundaries of pathogenic or spoilage microorganisms. In this point it was decided to conduct preliminary experimental trials which will further assist in the microbiological trials conducted in the next task on deli-type salads.
Study 1: To stablish the SOPs for the determination of the rheological properties of mayonnaise and mayonnaise-based salad dressing. Five different samples were tested having as variables the size of oil droplets, which were produced by altering the rounds per minute of the Coruma type homogenizer used for their production. The macroscopical parameter for checking the viscosity of the samples is he Plummit number (the distance that an arrow covers when falling into the emulsion from fixed height) used from the industries. The Plummit number of the samples showed good correlation with the maximum force needed for a Texture Analyzer probe to penetrate into the emulsion at a predefined distance.
Study 2: To evaluate the effect of different amounts of potato cubes added to mayonnaise or salad dressing base on the pH of the homogenous final product during storage. Seven combinations of mayonnaise + potato or salad dressing + potato (6:0; 5:1; 4:2; 3:3; 2:4; 1:5; 0:6) were prepared and stored at room temperature. The results showed that the pH of both products increased with the addition of potato and these changes followed linear correlation with the concentration of the product in the mixture. Following storage at room temperature, the pH of mayonnaise + potato mixture increased within 5 hours and then remained almost unchanged for the rest of the storage (20 hours). In contrast, this phenomenon was less evident in the salad-dressing base + potato mixture.
Moreover, CAMPDEN BRI has conducted shelf life trials on the three leafy salad products that have been chosen for the case studies. In total thirteen ingredients (Lollo Rosso, Escarole, Endive, Radicchio, Carrot, Red cabbage, Red peppers, Red onion, Mixed peppers, Sweetcorn) and final products (Leafy salad, Sweet and crisp, Mediterranean Bowl) were analysed. These ingredients and products were evaluated for yeasts and moulds, TVC, and Enterobacteriaceae over an 8 day life under MAP 15% CO2: 5% O2: 80% N2 and at 5oC in order to see whether:
a) There is much difference in the microbial levels for the 13 sample types and hence much difference in their assigned shelf-life.
b) To see whether the life of the finished products is influenced or defined by the life of the most sensitive ingredient.
Macroscopic inspection for moisture changes (products drying out or going watery) and browning of the products was also performed. There were definitely differences between the ingredients in terms of visual quality. We will probably base shelf-life on the TVC as the SOPHY partner BRYANS SALADS have their own in-house criteria to use. The action level of 107 CFU/g that Bryans use criterion was also used as spoilage level. In addition to shelf life studies for all the products from BRYANS SALADS, further microbiology and quality work has been conducted with the four leaf mixed salad. The data from this study showed as dominant flora Pseudomonas and yeasts, however, Enterobacteriaceae and lactic acid bacteria are also present in high level. The generated data of this study provide useful information about the microbial behavior on several types of vegetable ingredients that are used in mixed salads. By these means, the spoilage pattern of a mixed salad could be predicted by identifying the most sensitive ingredient.
The data collected in this task may be an info source for explaining future model deviations from real data. In addition these data will pinpoint potential hidden factors that compromise model performance and address the needs for new data generation.
Based on the conclusions and suggestions by the report on missing data, discussions were initiated on the needs and experimental designs for additional data generation for development new predictive models. The experiments that were discussed and designed are on the description of the shelf-life of leafy salads and the quality deterioration and safety of deli-type salads (mayonnaise and mayonnaise based dressings).
(a) Leafy salads: Microbial spoilage of fresh cut salads
With regards to the spoilage phenomenon of leafy salads, the AGRICULTURAL UNIVERSITY OF ATHENS and the NATIONAL TECHNICAL UNIVERSITY OF ATHENS examined different leafy salads in order to: (i) evaluate the effect of storage temperature and/or producer variability on the spoilage of commercial fresh cut-salads, (ii) to identify the responsible agents for the spoilage of different fresh vegetables, (iii) to evaluate the potential effect of the initial gas composition during packaging of fresh-cut salads on the selectivity of the specific spoilage microorganisms (i.e. lactic acid bacteria, pseudomonads) and (iv) to identify the O2/CO2 ratio threshold that may alter the spoilage pattern of the products.
Types of salads: Rocket salad (two batches), Iceberg lettuce, Romaine lettuce
Storage temperature: 0, 5, 10, 15oC and dynamic conditions (8 hours at 4oC, 8 hours at 12oC and 8 hours at 8oC)
O2 / CO2 ratio in MAP: O2 / CO2 ratio in MAP: 0% O2/20% CO2, 5% O2/15% CO2, 10% O2/10% CO2, 15% O2/5% CO2, 20% O2/0% CO2. The use of air perforated film for the packaging of the products was also examined.
Conducted analyses: Microbiological (total microbial flora, enterobacteria, lactic acid bacteria, pseudomonads, yeasts), burst strength, colour, sensory evaluation (colour, odour, texture, aroma), enzymatic activity, vitamin C degradation and, O2 / CO2 ratio in MAP.
The main results of the aforementioned microbiological studies can be summarized to the following: i) The dominant flora was Pseudomonas spp., regardless of temperature/ gas composition combination and type of fresh cut salad, possibly due to their high initial population compared to the other microbial groups, ii) Even though lactic acid bacteria had low initial level, they played significant role to the microbial spoilage due to their high growth dynamics at all assays, iii) Overall, the factor that was responsible for the quality deterioration during storage differed among the tested fresh cut-salads and thus, the spoilage profile significantly differed among the products.
(b) Deli salads: (1) Shelf life of Farmer salad, North sea shrimp salad and German sausage meat salad
Type of salads: Farmer salad, mayonnaise based, contains vegetables, with preservatives (sodium benzoate, potassium sorbate); North sea shrimp salad, mayonnaise based, contains shrimps (Crangon crangon), with preservatives (sodium benzoate, potassium sorbate); German sausage meat salad, mayonnaise based, contains sausage meat and gherkins, without preservatives
pH adjustment: Farmer salad: 3.95 – 3.60 – 3.35 – 3.20 – 3.10; North sea shrimp salad: 5.00 – 4.70 – 4.40 – 4.10 – 3.80 – 3.50; German sausage meat salad: 4.40 – 4.20 – 4.00 – 3.80 – 3.60 – 3.40
Storage temperature: 4°C, 8°C, 15°C, 25°C
Total viable counts (TVC), lactic acid bacteria (LAB), and yeasts and moulds.
LAB were the dominant microorganism at all studied experimental assays and all types of salads. The growth dynamics of TVC was increased as the initial pH and the storage temperature was increased, regardless of type of salad. With regards to pH, in Farmer and North sea shrimp salads the pH was decreased depending on the storage temperature, while in German sausage meat salad the initial pH remained stable during storage.

Deli salads: (2) Microbiological spoilage of potato and chicken salad
Experimental design: Mayonnaise-based deli salads (potato salad, chicken salad and cream-dressing) were provided by KALAS; acidulant: acetic acid, lactic acid; pH: 3.60 3.90 4.10 4.40; storage temperature: 0, 5, 10, 15°C; Potato/chicken/cream ratio: 1/1 (realistic scenario), 2/1 (extreme scenario); Preservative (sorbic, 0.11% w/w); Average weight of cream salad in samples: 40 g; Microbiological (total microbial flora, lactic acid bacteria, pseudomonads, yeasts).
The objectives of these studies were to evaluate: (i) The effect of different homogenization levels (mayonnaise) or thickener concentrations (salad dressing base), (ii) The effect of different pH levels adjusted with different organic acids (i.e. acetic acid, citric acid or sodium bisulfate), (iii) The effect of 0.1% sodium sorbate (preservative) on the survival of pathogenic bacteria or the spoilage rate of the products.
The main results of the aforementioned studies can be summarized to the following:
Potato deli salad: The results revealed that the spoilage profile significantly differed among the products. Microbial growth was observed in sorbic free potato salads with lactic acid as acidulant. The other formulations resulted in very slow microbial growth and physicochemical changes.
Chicken deli salad: The spoilage of chicken salad seems to be primarily affected by the microbial activity of lactic acid bacteria, causing changes in the sensory properties of the product (i.e. acidification, off-odors). The addition of sorbic acid resulted in significant stability in terms of microbial growth and sensory changes at all storage temperatures. PH values decreased slightly during refrigerated storage with no significant differences.
Deli salads: (3) Survival of Salmonella spp. and L. monocytogenes in potato and chicken salad
Challenge test # 1: Survival of Salmonella spp. in potato salad, as affected by: the initial pH of the cream salad, the concentration of the particulate (potato or chicken), the type of the acidulant, and the presence of preservative.
Acidulant: Acetic acid, lactic acid; Initial pH: 3.60 3.90 4.10 4.40; Potato concentration: 0%, 33.3%, 50%, 66.7%, 75%; Preservative: With or without 0.1% sodium sorbate; Storage temperature: 5oC.
The initial pH of the cream salad affected the survival of the pathogen in samples without potato, with samples of pH 3.6 eliminating the pathogen within 24 hours. Depending on the concentration of the added potato, the pH between two particles increased by 0.1 to 0.5. This increase occurred during the first 24 - 36 hours, while no significant changes were observed during further storage. Salmonella spp. survived 5.8 log CFU/g in samples with 1/1 ratio of potato/cream salad, where the pH was adjusted at 3.98 after 24 hours of storage at 5oC. However, the survival of the pathogen was limited at 4.6 log CFU/g in samples without potato and pH 3.94 suggesting that the viability of Salmonella is affected not only by the pH of the occurring environment, but also by the concentration of the potato that may decrease the acidity of the final product.
Challenge test # 2: Survival of Salmonella spp. in chicken salad, as affected by: the initial pH of the cream salad, the concentration of the particulate (potato or chicken), the type of the acidulant, and the presence of preservative.
Acidulant: Lactic acid, Acetic acid; Initial pH: 3.60 3.90 4.10 4.40; Chicken concentration: 0%, 33.3%, 50%, 66.7%, 75% (lactic acid); 0%, 10%, 20%, 30%, 50% (acetic acid); Preservative: With or without 0.1% sodium sorbate; Storage temperature: 5oC.
In contrast to potato salad, the pH in chicken salad increased at higher levels, resulting in lower reductions of the pathogen. This took place even in products with low concentration of chicken (i.e. 10-25%). Acedic acid caused higher reduction of the pathogen compared to the respective ones with lactic acid, especially in the presence of the preservative.l
Challenge test # 3: Growth/No growth interface of L. monocytogenes in chicken salad, as affected by: the initial pH of the cream salad, the concentration of the particulate (chicken), the type of the acidulant, and the presence of preservative.
Acidulant: Acetic acid, Lactic acid; Initial pH: 3.60 3.90 4.10 4.40; Chicken concentration: 0%, 10%, 20%, 30%, 40%, 50%; Preservative: With or without 0.3% sodium sorbate; Storage temperature: 5oC.
The growth potential of L. monocytogenes in the chicken salad of different formulations followed the order: lactic acid without preservative> lactic acid with preservative> acetic acid without preservative> acetic acid with preservative.
Deli salads (4): Impact of dynamic pH change over time on Salmonella serovars.
The experimental conditions of the above objective are described below:
Microorganisms: single strain (S. Typhimurium LT2), Cocktail mixture (S. Typhimurium PS1, S. Typhimurium PS2, S. Agona, S. Reading and S. Enteritidis); Temperatures: Ambient (22ºC) and refrigeration (6ºC); pH change: Instant and gradient (over a period of 6h); pH range: 6→3; Acids: Hydrochloric (HCl) and acetic (CH3COOH); Microbial counts: Culture on media (TSA) and Flow cytometry.
The main results of the aforementioned studies can be summarized to the following: i)The gradient pH change increased the inactivation rate of Salmonella (HCl), ii) Lower temperature reduced the inactivation rate of Salmonella, iii) Acetic acid had a greater effect on the inactivation of Salmonella than HCl, iv) The cocktail mixture of Salmonella found to be more resistant to pH changes than the laboratory strain (S. Typhimurium LT2), In the case of gradient pH change flow cytometry revealed changes in the ratios of the healthy, injured and dead cells at different time intervals
After the data generation for development of new kinetic and probabilistic models for microbial survival/growth data was generated for microorganism-independent quality decay models.
Leafy salads (1): Quality decay parameters of fresh cut salads
In this Task the quality parameters were explored that could be used to describe the effect of storage temperature and modified atmosphere packaging conditions on rocket, romaine lettuce and iceberg lettuce accordingly to the experimental design described in the Task before. The quality parameters studied involved: changes in appearance, texture, enzymatic reactions, color, aroma, flavor, texture, and vitamin C content.
I. Colour in leafy salads: Quantitation of the colour change was based on measurement of CIELab values. Five pieces of salad leaves, representative of the products, were measured.
The main results are summarized to: i) The storage temperature significantly affected luminosity and increasing temperature decreased luminosity values, ii) A significant increase in a, indicating a shift from greenness to redness was observed in all the samples during storage. Parameter a is a colour parameter related to browning and to the breakdown of chlorophyll, iii) Total colour difference (DE values) increased with storage time approaching an upper asymptote (sigmoid response), iv) Based on DE values, the colour change/deterioration was more intense for romaine lettuce presenting the highest DE values, while rocket samples showed the lowest rates of colour deterioration as well as the lowest DE values, v) The increase of storage temperature caused increase of colour deterioration rates, as expected.
II. Sensory properties of leafy salads: Ten trained panelists working in the NATIONAL TECHNICAL UNIVERSITY OF ATHENS were chosen for the assessment of the sensory attributes. Overall visual acceptability, odour, texture, taste, freshness and overall impression were rated by using a nine point scale where 1 = absent or poor characteristic, 9 = too intense or excellent). During sensory evaluation test, panelists were also asked for the acceptability of the leafy salad samples (Accepted or Not accepted). Rocket samples were not acceptable by the panelists after 8 (5% O2, 15% CO2, 80% N2) days of storage at the temperature of 5°C. Acceptability of iceberg lettuce samples were different for different packages, 10 (5% O2, 15% CO2, 80% N2)- 20 (5% O2, 15% CO2, 80% N2) < 5 (aerobic) days of storage at 5°C, and 6 (5% O2, 15% CO2, 80% N2)-10 (5% O2, 15% CO2, 80% N2) days of storage at 15°C. The romaine lettuce samples presented 9 days of storage at 5°C and 2 days at 15°C, not affected by the packaging conditions.
III. Vitamin C (L-ascorbic acid) in leafy salads: Vitamin C (L-ascorbic acid) was determined using a HPLC (Giannakourou and Taoukis, 2003). Rocket salad leaves contained high levels of vitamin C. The non-homogeneous nature of raw material is reflected in the significant variability of the initial nutritional value. The iceberg lettuce did not present significant levels of vitamin C. Vitamin C loss (e.g. 50-70%) have been used for shelf life evaluation. The acceptability level of vitamin C loss is recommended to be chosen in accordance with the level of sensory acceptability. The vitamin C loss for rocket salad samples (all types of packages) was found to be 50% (of the initial value), approximately, after 8 days of storage at the temperature of 5°C.
IV. Changes in texture: Burst strength values were measured against storage time for each storage temperature for all threes studied salads at all studied gas compositions (CO2 percentages). Τhe burst strength (=maximum force for rupture) was found to decrease when plotted with time for every storage temperature studied. For each storage temperature, the burst strength was measured at two different points of non-injured romaine lettuce or rocket leaves at ambient temperature for at least five different leaves. The puncture strength was also measured for the iceberg. The main results are summarized to the following: i) Higher storage temperatures resulted in lower burst strength values when compared with corresponding values of leafy salads stored at lower temperatures, ii) In contrary to these results, the puncture strength for the iceberg was found not to be significantly affected by storage time, temperature and gas composition (CO2 percentage), iii) Burst strength of all rocket and romaine samples was decreased vs storage time, while for samples stored at higher temperatures, the decrease was faster than for those stored at lower temperatures, iv) No significant effect of storage time and temperature, as well as gas composition on the observed burst strength values of iceberg samples.
V. Enzymatic reactions: Polyphenol oxidase (PPO), Quinone, and L-Phenylalanine ammonia-lyase extraction were performed according to previous literature (Fang et al., 2006; Degl’Innocenti et al., 2007; Zhan et al., 2012). In rocket, the enzymatic activity of PPO decreased during storage at all assays, while in iceberg PPOs’ activity increased especially as oxygen was increased. Quinone activity remained stable during storage of rocket, while in iceberg and romaine revealed significant increase.
VI. Multi- and hyper- spectral non invasive techniques: Multispectral image analysis and Fourier transform infrared spectroscopy are ranked among advanced sensing technologies that can effectively and efficiently inspect products quality and safety. These type of information is of particular importance since it is known that chemical properties of foods are dependent on the chemical distribution within the samples. The results of the multispectral analysis revealed that this technique may adequately describe the differences on the chemical composition e.g. yellow leaves (470-700 nm), sprinklyness (surface moisture) (780-940 nm), or browning on the surface of the vegetables, as it is affected by the storage temperature, time or the modified atmosphere gas composition. For example, in iceberg salad, oxygen increase had a significant effect on the % reflectance of the samples, indicating the chemical alteration due to browning, while in rocket the increase of CO2 showed higher reflectances near the 780-940 nm which have been related to increased surface moisture. With regards to FT-IR, the retrieved spectrum represents the molecular absorption and transmission, creating a molecular fingerprint of the sample. In particular, the retrieved spectrum of romaine and iceberg lettuce appeared to be affected by both the gas composition of the package, the storage time and temperature, compared with a control sample. Further identification of the particular chemical components (peaks of the spectrum) will possibly provide information that could associate these changes with the microbial/enzymatic activity.
Leafy salads (2): Effect of various storage and packaging conditions on the texture/firmness and the appearance of fresh-cut salads
THE UNIVERSITY OF BIRMINGHAM and CAMPDEN BRI developed a new method for texture analysis and surface roughness determination under various storage and packaging conditions of fresh-cut salads. These studies were performed in order to clarify the possibility of grouping fresh vegetables according to their surface roughness profile. Combining such information with the generated data of Task 3.2 and 3.3 will facilitate the development of unified (generic-like) models for the prediction of the microbial behavior on whole groups (i.e. rough, semi-rough, smooth) leafy vegetables. The samples used in these studies were from Bryans leafy salads which each bag (110g) contained approximately 38% Lollo Rosso, 34% Escarole, 17% Endive and 11% Radicchio. The shelf life of fresh-cut salads is 7 days.
Study 1: As a pilot study, the shelf life of five types of fresh-cut salad (Lollo Rosso, Escarole, Radicchio, Endive and Mixed Leafy Salad) packaged under 5% O2, 15% CO2 and 80% N2 and stored at 5˚C for 7 days was evaluated for the development of a quality model. Three measurements were implemented during storage (on days 0, 4, and 11). For each sample, photos were taken and the dimension and weight was measured during storage through an interferometer (optical technique that uses of the principle of superposition to combine waves aiming to construct 3D surface maps). Roughness and mechanical tests were also implemented. The elasticity was measured for each type of vegetable separately (20 g). The results from those pilot studies showed that differences in texture could be detected as the storage time, and thus the degradation, under isothermal conditions increased.
Study 2: To quantify differences in texture of different types of vegetables during various storage and packaging conditions. The tested O2/CO2/N2 ratios and the storage temperatures are summarised in the Task before. The modified atmosphere packaging conditions studied so far were 5% O2, 15% CO2 and 80% N2 and the salads were stored at a 5˚C controlled temperature cabinet for 7 days. In particular, the elasticity of fresh-cut mixed salads (BYANS leafy salad) was measured. Samples were analysed after 0, 4, 5, 6 and 7 days of storage in triplicate. The main results are summarized to: i) Salad leaves remained harder during storage at a gas composition with less oxygen and higher carbon dioxide (5%O2-15%CO2-80%N2) and at storage temperatures 0˚C and/or 5˚C, ii) An increase of the storage temperature lead to an increase of the deterioration process, iii) Endive leaves tend to be less sensitive (up to 0.50%), than Lollo Rosso MAP conditions and storage temperatures. The deterioration area tended to increase significantly at 10oC, and in all cases, salad leaves seem to be better conserved when stored at 0oC or 5oC, iv) The combination texture and deterioration analysis lead to the conclusion that the best storage condition is the one with less oxygen and higher carbon dioxide (5%O2-15%CO2-80%N2) and at storage temperatures 0˚C and/or 5˚C, v) The results showed that this method can be used to quantify the effect of the storage conditions on the texture of the samples and will be further used to examine potential differences between the conditions examined herein against other combinations of gas compositions (O2, CO2 and N2) as well as other storage temperatures e.g. 8˚C proposed by the SOPHY partners.
Deli salads: Quality decay parameters of potato and chicken salad
NATIONAL TECHNICAL UNIVERSITY OF ATHENS explored the quality parameters such as colour, sensory evaluation (colour, odour, texture, and aroma) that could be used to describe the effect of storage temperature, potato/chicken/cream ratio, type of acidulant, initial pH, presence or absence of preservative on potato salad, chicken salad and cream-dressing, accordingly to the experimental design described in Task 3.3. Mayonnaise-based deli salads were provided by KALAS and shelf life experiments were implemented. Quantitation of the colour change was based on measurement of CIELab values. Rating of sensory attributes was carried out using a nine point scale (0 = non-existent – imperceptible or poor characteristic, 9 = too intense or excellent). The results revealed that the spoilage profile significantly differed among the products. The spoilage of chicken salad seemed to be primarily affected by the microbial activity of lactic acid bacteria, causing changes in the sensory properties of the product (i.e. acidification, off-odours). Total colour difference (DE value) increased during storage, with the highest temperatures (i.e. 10 and 15⁰C) showing the highest DE values. Sorbic acid resulted in significant stability in terms of appearance and overall sensory impression at all storage temperatures. At all temperatures studied, the time of sensory rejection (score 6 for overall impression) coincided with an average LAB count of 8.5 logcfu/g for sorbic free samples and lactic acid as acidulant, and for samples with 0.11% w/w as preservative and lactic acid as acidulant (initial pH=3.6).

WP4
A detailed meta-analysis of existing/available data on the microbial growth of spoilage or pathogenic bacteria, as well as in quality decay parameters in fresh produce (focusing on leafy salads) was performed, in order to identify areas where microbial behavior (already described) is highly variable or uncertain and to characterize the factors affecting the variance of the microbial responses (growth and inactivation rates). In addition, an attempt to identify and characterize the factors affecting the quality and thus the shelf-life of vegetables was made. The experimental areas which were investigated for the existence of uncertainty in microbial responses, were associated with: (i) the variables which were identified as critical for being used in the SOPHY software, (ii) the expected use of the software (e.g. offering characterization of safety and spoilage level, decision support, etc.) and the potential end-users (e.g. SMEs). Available data consisted of Log CFU/g, as well as growth rates of various microorganisms in fresh produce (focusing on leafy salads) published in peer review scientific journals, text books and data publicly available in scientific databases, such as ComBase. In addition, quality decay rates derived from the fit of zero and first order models published in peer review scientific journals were also used for the meta-analysis of the quality decay of fresh produce. The review of literature data was carried out using the valuable tool of SOPHY database that has been developed within SOPHY project, as well as records from the ComBase. The systematically collected records in the SOPHY database facilitated a more effective method of retrieving data concerning the mathematical models and the respective parameters of microbial growth in fresh cut salads. All analyses were performed in Microsoft office excel (Microsoft, (2007). Microsoft excel [computer software]. Redmond, Washington: Microsoft) and JMP 8 statistical software (SAS Institute Cary NC USA).
In total 1059 entries for growth data, 93 entries for inactivation data and 1809 entries for quality data from both ComBase and SOPHY Databases were gathered and analysed. The growth and inactivation rates of pathogenic and/or spoilage microorganisms in fresh produce showed that they follow an Arrhenius type relationship. Temperature governs the growth and inactivation rates. Other intrinsic factors such as pH and aw seemed to have an effect on the rates as they explained a percentage of the observed variability of the Arrhenius relationship. However there is no available information for these factors on their effect on the growth rates of spoilage microorganisms. Bacterial species and products (vegetables) found to be relatively independent of the growth/inactivation rates. In cases where information about intrinsic factors such as pH and aw were not available, a distribution of the growth rates at specific temperature ranges (e.g. 3-6°C which can be considered as the refrigeration range of temperatures used commercially) showed that the growth rates of different classes of bacteria follow a Log-Normal distribution with similar scale and shape parameters. This also suggests that temperature is the factor which governs the growth rates independently of the bacterial species. This might suggest that a unified model can be constructed to obtain kinetic parameters if essential information are known (such as temperature and pH), regardless the type of microorganism or product. Therefore, more research should be undertaken to fill in the gaps that exist in the literature in order to be able to model the relationships of the kinetic parameters and explain the observed variability. These results can help in decision making for a wide range of food safety questions.
Regarding quality decay parameters, the sensory and instrumental methods resulted in similar distributions of quality decay rates in most of the cases. However, in some cases the instrumental method resulted in a wider distribution of rates with higher sometimes reaction rates. This might be a result from the actual method itself. Sensory methods use a trained panel of people who judge a product for several quality indices such as visual quality, colour, texture, aroma, flavour or appearance and thus consists a direct but subjective/inaccurate method. On the other hand, instrumental methods use instruments to examine several quality indices mainly colour and texture. Instrumental methods are indirect methods of examining quality but are more accurate. The selection of the quality index in combination with the product and the processing and packaging conditions found to be the main parameters which will determine the actual shelf-life of the product if the temperature has been controlled properly throughout the cold chain (production, distribution and storage). It can finally be suggested that both instrumental and sensory methods should be used in conjunction with each other in order to give the desired objective. Instrumental methods can give a more accurate result while the sensory method is more relevant to consumer’s acceptability.
Based on the findings of the data generation work package thirty nine new kinetic or Growth/No growth models for the behaviour of pathogenic bacteria on leafy or deli-type RTE foods were developed. This Task was linked to five independent experimental trials, from which the obtained data were used as the basis of the modelling procedure conducted in development of quality and safety models.
Trial #1: Development of mathematical models for the survival of Salmonella, Listeria monocytogenes and E.coli during decontamination treatments of leafy salads.
The three pathogens were exposed to different concentrations of sodium hypochlorite or paracetic acid for different washing times. As a further step, the behavior (reduction rate) of each pathogen was described as a function of exposure time and the concentration of the sanitizer, using a typical polynomial equation: reduction rate =a+b*x+c*y+d*x2+e*y2, where x is the exposure time, y the concentration of the sanitizers and a, b, c, d, e are the model estimates, as derived from the regression analysis of the obtained reduction rates and the corresponding exposure times and concentrations of the sanitizers.
Trial #2: Development of mathematical models for the impact of dynamic pH change over time on Salmonella serovars.
A model system (broth) was inoculated with Salmonella serovars (single strain and cocktail mixture) and was subjected to an instant and gradient pH over time (during a 6 h period). The pH adjusted between pH 6 to pH 3 using HCl and acetic acid at ambient (22°C) and chilled temperatures (6°C). The survivors of the pathogen were fitted by four non-linear models (convex, linear, monophasic logistic and Weibull) in order to describe the survival kinetics of the pathogen with time (Figure 2). All models had a good fit to the data (low RMSE) but the Weibull model found to be the model which had the best fit in most experiments carried out.
Trial #3-4: Development of mathematical models for the survival of Salmonella in potato and chicken salad
Eight independent models were developed for the prediction of Salmonella reduction in potato or chicken salad acidified with (i) acetic acid without sodium sorbate, (ii) acetic acid with 0.3% sodium sorbate, (iii) lactic acid without sodium sorbate, (iv) lactic acid with 0.3% sodium sorbate. The log-transformed populations of the pathogen were fitted by the Weibull model in order to calculate the delta value (i.e. the time for the first logarithmic reduction), as it was affected by the type of the acidulant, the initial pH of the cream base and the concentration of the potato. The delta values of the pathogen were fitted by a polynomial equation in order to describe the effect of the initial pH of cream salad and the concentration of potato/chicken on the delta value of Salmonella spp. (Table 2). All polynomial models were able to describe the survival kinetics of Salmonella during storage of the products, especially in products acidified with acetic acid and the addition of 0.1% sodium sorbate. However, due to the high increase of the initial pH of chicken salad after the addition of the chicken, the goodness of fit of the secondary models in this product were lower compared with potato salad.
Trial #5: Development of a Growth/No growth model for L. monocytogenes in chicken salad
The objective of this trial was to develop a mathematical model for the prediction of the G/NG interface of L. monocytogenes in chicken salad. As described before, a full factorial experimental design was prepared in order to include a wide range of cream salads with different initial pH and different concentrations of chicken. Growth was indicated when the observed log-increase after 30 days at 5oC was two times higher than the standard deviation of counts on time 0. Growth or no growth was scored as 1 or 0 values, respectively. The derived data were fitted to a logistic regression model [Logit(p)=a0+a1*pHin+a2*(pHin-pHmean)2+a3*% chicken+a4*(% chicken-% chickenmean)2 +a5 *pHin *% chicken], using JMP software (Table3). The developed models were able to accurately describe the G/NG interface of L. monocytogenes in chicken salad, especially in those acidified with acetic acid and formulated with 0.3% sodium sorbate. In products with lactic acid and without sodium sorbate, the G/NG interface of the pathogen was affected by the increased activity of the indigenous microbiota and, therefore this models showed higher uncertainty compared with the models in products with acetic acid and/or 0.3% sodium sorbate.
A great amount of microbiological data was collected from two large-scale, independent trials on the spoilage of fresh cut and deli-type salads. These data constituted the base for the development of mathematical models for the prediction of the spoilage of three fresh-cut and two deli-type final products.
Trial #1: Modelling the spoilage of fresh cut salads
Samples of Romaine lettuce, Iceberg lettuce and Rocket (two batches) were stored under different temperatures and MAP conditions (for more detailed data, please refer to D.3.3). Results showed that pseudomonads and lactic acid bacteria were found to be responsible for the microbiological deterioration of the salads quality. Therefore, the generated data (log CFU/g) for these microorganisms were primarily fitted by the Baranyi model to obtain the corresponding kinetic parameters (μmax and lag time). As a further step, three secondary models were developed and evaluated for their ability to accurately predict the growth of these two microorganisms in the four examined products, as a function of the storage temperature and the CO2 concentration. In particular a polynomial, the Arrhenius and the Belehradek models were used to describe the maximum specific growth rate of pseudomonads and lactic acid bacteria.
Trial #2: Modelling the spoilage of deli-type salads
Mayonnaise-based deli salads were provided by KALAS and shelf life experiments were designed, in order to describe the microbial spoilage of chicken and potato salad as it is affected by the type of the acidulant (acetic acid, lactic acid), the product pH (four different levels), the storage temperature (4 levels), the presence or absence of preservative (0.3% sodium sorbate) and the concentration of the particulate (2 levels).
Due to the very slow changes in microbial counts during storage of potato salads with acetic acid (with or without preservative) and lactic acid with preservative, there were no modelable data for quality deterioration in these products. In contrast, the microbial growth was modelled in sorbic free potato salads with lactic acid as acidulant, where the effect of temperature on LAB and yeast and moulds growth rates was adequately described by Arrhenius equation. Similarly, the effect of storage temperature on the growth rates of LAB and yeasts was also described by the Arrhenius equation. In a further step, secondary (polynomial) models were developed to predict the microbial growth in chicken salad, as a function of storage temperature and initial pH.
Data generated in the data generation work package were used to develop mathematical models describing the kinetics of non-microbiological, quality parameters during leafy salads deterioration as a function of storage time, storage temperature and packaging gas composition. In addition, a food model system was developed to visualize the pH gradient in deli-type salads after the addition of the particulates.

Trial #1: Modelling the changes of non-microbiological indices in leafy salads
The effect of storage temperature on phenylalanine ammonia lyase (PAL) activity, quinone content, ascorbic acid (Vitamin C), texture (burst strength) and sensory characteristics of romaine lettuce, iceberg lettuce and rocket was investigated. Firstly, PAL, quinone, sensory and vitamin C was described as a function of storage time using an exponential equation (first order kinetics), while for texture, burst strength was correlated with time using a typical linear equation. Temperature dependency of all non-microbiological indices was further described by the Arrhenius equation (Figure 3, 4). All models could adequately describe the evolution of the quality indices during storage of the leafy salads, as it was also confirmed by the correlation of the predicted and the observed (experimental) values.
In parallel, a wide range of spectrafrom multispectral imaging analysis which corresponds to the potential different chemical conditions that occur on the surface of romaine lettuce, iceberg and rocket salad during storage were collected in WP3. Since these data showed that there is a correlation between the % reflectance and the storage time of the leafy salads, this factor was linearly modelled as a function of storage time. Three wavelengths were chosen (570, 890, 970 nm) as they may illustrate three major quality factors (yellow colour on leaves, sprinklyness- surface moisture and browning, respectively). The rate of reflectance changes was further modelled as a function of storage temperature and the CO2 concentration of the MAP.
Trial #2: Modelling the changes of non-microbiological indices in deli-type salads
The same modelling approach as in leafy salads was followed in order to mathematically describe the sensory evaluation of chicken salad as a function of the storage temperature. In particular, the sensory perception for colour, texture, taste and general impression of chicken salad expressed by the scores given for the respective quality parameter vs. storage time was primarily described by linear equations. The effect of temperature on sensory scoring (general impression) change rates was further described through the Arrhenius equation.
Trial #3: Modelling the pH gradient in deli-type salads
An Agar gel (pH=5.4) was prepared and used as a model system to monitor the pH gradient around the particulates in deli salads. Bromophenol blue was added to the system to allow the visualisation of the pH gradient. Commercial mayonnaise (pH 3.5) was added in the centre of the system (Picture 1). The model system could be used to visualise the pH interaction between 2 components using a pH indicator, and the pH gradient could be recorded and quantified with 0.2 pH precision. Following the correlation between Hue value and pH for the bromophenol blue, the interaction between mayonnaise and agar gel was observed by the gradual change of pH at 5, 10, 30 and 60 min. At each time, the Hue map distribution obtained after image processing was used to visualize the pH gradient around the mayonnaise.

WP5
The key objectives of the software development have been twofold: a) Providing a user-friendly environment where modelling specialists can provide models to the software and describe the intention of models to the system; b) Providing SME users with an environment through which they can model their production chain (including storage / distribution steps after production) and predict the quality and safety effects that their modelled production chain might have on a product should the chain be truly implemented. Key to this is the ability for SME users to experiment with processing parameters throughout the modelled chain, being able to make adjustments to processing conditions in order to see the effect those changes may have on quality and safety.
Objectives and challenges in providing an environment for scientist:
During early discussions, it has been clear that it is very desirable that the SOPHY software has an understanding as to the intention of any quality or safety model introduced to the system. By understanding, the software developer mean that the system has sufficient knowledge about every model submitted to the system to know when that model is applicable for being called upon to solve a particular problem, with a problem being a prediction that needs to be made on behalf of an SME user. The scope of such knowledge to be present for each model is a collection of model-level characteristics that can be defined in the software as: The product(s) or ingredient(s) for which a model is suited to generating predictions (e.g. Iceberg Lettuce); the quality or safety index to which a model is suited (e.g. Lactic Acid Bacteria); the industrial process to which a model is suited (e.g. Slicing). The combination of these characteristics defines the role a model has in the system – in other words the conditions under which a model may be deemed as being suitable for selection in determining a prediction on behalf of an SME user. It should be noted that a key objective of the scientific community within the SOPHY project is to introduce highly product, process, quality & safety index-specific models in order to target high accuracy in terms of predictions (the models are not generic). The key objective of the meta-data approach is to raise the level of abstraction surrounding the selection of the correct models for a specific production scenario such that the end user is not confronted with making the selection of models themselves. A key benefit of the efforts to raise the level of abstraction in terms of model selection is that an end SME user does not, themselves, require an intimate awareness of microbiological models (and the selection thereof) in order to carry out quality and safety predictions.
In order to cater for a mechanism which can perform such model selection on behalf of end users, a meta-data approach has been required at the model level. The SOPHY database was structured in such a way as to allow the aforementioned key characteristics of a model to be stored as a model definition. Additionally the user interface developed for modelling experts is presented in such a way as to allow experts to easily select the key characteristics that apply to a model via simple drop-down lists. The selection of just three drop-down lists is sufficient to describe the model via meta-data so that the runtime of the SOPHY software knows when a model is suitable for selection.
A collection of meta-data which describes a model’s role in the system is of course not enough to generate predictions; models must also undertake the aspect of carrying out calculations – calculations which are driven by a number of inputs which represent real-world processing parameters (such as temperature, initial contamination). The meta-data stored against a model definition in the SOPHY software’s database therefore includes provision for the one-to-many storage of model inputs and model outputs that are required by a model in order to usefully generate a prediction. It should be noted that the level of meta-data storage concerned with model definitions relates to the types of inputs a model will require, and the types of outputs it will generate; it does not relate (at this stage) to the actual values that are provided to a model, since this is provided by an end SME user. Indeed the meta-data describing the types of inputs and outputs expected by a model is leveraged later by the SOPHY software to actually prompt an end SME user to provide processing parameters which are required by any given model. And since a model, via its meta-data, is product / process bound, the parameter prompts which are generated for a user are highly relevant.
Besides provision at the meta-data level for describing the role, as well as the inputs and outputs of model, a mechanism for carrying out the actual calculations necessary to form a prediction is also required. A couple of key objectives where identified, together with modelling experts, in terms of the calculation aspect of models: a) A model’s calculations should function accurately in the software – that is to say, in terms of calculations, a model must perform within the software in precisely the same way as it was developed during research; b) Models should be independently testable, ideally before they become part of the software – this is important because scientists need to independently verify the correctness of a model under scientific conditions before the model is a candidate for becoming part of the SOPHY software. This is a particularly hard challenge, since models are developed very much at the mathematical level, whilst software operates at the algorithmic level. The concern is, of course, that the calculations developed at the mathematical level are translated correctly in terms of instructions at the algorithmic level. What transpired quite early on in the project is that SOPHY RTDs constructs their mathematical models in terms of Excel sheets with a number of inter-related cells containing formulas. Whilst one approach towards software integration may be the step-wise decomposition of an Excel sheet (by manually following the relationships between all cells) and subsequently coding equivalent algorithms, it’s a process which is potentially very error prone, rather defeating the key objectives of accuracy and testability. It was therefore decided that in order to meet these key objectives, software would be developed within the SOPHY project which allows: a) Excel-based models to be submitted verbatim to the software – the exact same Excel model which is developed by modelling experts is directly used in calculations in the software; b) Excel models to be consulted by the software during runtime and manipulated in memory in such a way that parameters from an end SME user’s processing conditions could be injected into the model, with results conversely being extracted from the model and introduced back into the SOPHY software. In other words – verbatim Excel models could be used as if they were a software engine. A layer of software was therefore developed which could generically coordinate the process of injecting and extracting data to and from any arbitrary Excel model as a background service, rendering the whole process completely invisible to end users.
The combination of meta-data and a layer of software which effectively drives Excel models as if they were dedicated software engines provides for an extremely flexible mechanism under which many, many different models, each with potentially different inputs and outputs, can be added to the system simply by describing their characteristics and uploading a corresponding Excel model. In future the same mechanism could be expanded with other software layers for driving other “engines” such as models compiled through Matlab. Nevertheless, Excel has proved sufficient for the purposes of SOPHY’s modelling.
Every Excel model generates a prediction for a single index, such as Salmonella, which is modelled over time. Time is a variable which is always required in every model; every Excel model creates a 2D array of [time, predicted value] results. When an Excel model is loaded and consulted by the SOPHY software, two core things are extracted: a) A [predicted value] at time [t] where [t] is the duration of a process being modelled by an end SME user; b) The whole 2D array. The single scalar predicted value at time [t] is considered to be the output variable of the model and often used as the input value for a subsequent process modelled by an SME; the 2D array is used for generating a graph of results over a period of time which corresponds to the total duration of the process as expressed in input variable [t]. The SOPHY software uses mappings at the meta-data level to determine, for each model, where within the Excel sheet the [t] variable should be injected, and where the 2D array can be found. The result is that the SOPHY software can successfully extract a single scalar output value as well as a 2D array which is highly suited for creating a final graph of results.
Alongside meta-data for describing a model’s characteristics, additional meta-data can be supplied by a modelling expert to apply certain industry best practices via thresholds. Thresholds allow a modelling expert to examine any single scalar output generated from a model by comparing it to conditional criteria. As an example, if a model returned a log CFU/g value for a Salmonella prediction, a modelling expert can apply a condition to trigger a warning should that value exceed a certain log value. In this way, certain best practices can be built into the system to trigger warnings for SMEs when it is predicted that their processes exceed certain value.
In addition, modelling experts can also attach any number of pre-built information sheets to a model. The information sheets are pre-loaded into the software in the form of PDF documents. When a modelling expert attaches information sheets to a model, he/she is attaching model-relevant guidance which, when a model is selected as being relevant to an SMEs processes, will be displayed by clicking on a hyperlink.
Objectives and challenges in providing an environment for end SME users:
A key goal of the SOPHY software is to allow end SME users to experiment with product processing and product formulation in such a way that useful predictions can be generated about the quality and/or safety of the processes in question. In accordance with the software design documented at an early stage of the SOPHY project, the software allows users to create multiple processing scenarios. A scenario begins as a virtually blank canvas with one vertical block available for adding processes. In the software, we refer to a vertical block as a swimlane; multiple successive swimlanes can be added to a scenario, with each swimlane being used to organize the sequence of processes that take place in an overall scenario. For example, the very first processes that occur in a scenario are organized in the first swimlane, the next successive processes that occur in sequence are organized into the second swimlane, and so on. In this way, the whole sequence of a number of processes can be defined within a scenario. A single swimlane can house multiple processes, each of which are related to each other simply by sequence (e.g. the first thing that was done with the potato in this scenario is … + the first thing that was done with the carrot in this scenario is … … … are processes that all belong in the first swimlane).
When any process is added to a scenario, a user is prompted to provide information regarding the characteristics of the process: a name must be provided for the process; one or more products / ingredients must be selected for the process; an activity must be selected for the process (e.g. slicing). A key objective of the SOPHY software is to alleviate the user from having to select relevant scientific models which are applicable to a given process and its characteristics, since many end SME users will likely not have the intimate knowledge required to select the correct models relevant to a process. The SOPHY software achieves this objective by leveraging the set of meta-data that a modelling expert has applied to each model in the system. Therefore, when a user selects a given set of products, along with an activity, a database look-up takes place against the meta-data that has been provided. This look-up yields models that are highly relevant to the given process because the meta-data at the model level says so and has been entered by a modelling expert. In other words, the system is able to leverage the expert knowledge of the modelling expert in aiding model selection. Models are selected for all possible available quality / safety indexes where a match can be made on the combination of: a) ALL products / ingredients selected in a process; b) the activity of a process.
When model selection takes place, the meta-data stored against models is yet further leveraged to supply the end SME user with a set of highly relevant parameters that can be entered. The user does not have to consider themselves which parameters may be relevant; the system does this for them by leveraging the meta-data at the model level which describes the parameters that a model needs in order to a make a given prediction. Whilst the parameters that are required vary according to each model selected, one parameter in particular is always required, namely Process Duration. Since the model selection process often returns many models for any given single process in a scenario, and since many models often share the same parameters (such as pH), the system must be intelligent enough to de-duplicate the parameters that are presented to the user. Model selection and parameter generation has been widely seen in the consortium as a very valuable goal to achieve and has also been widely welcomed by industrial users since it makes the software easier to use.
As indicated above, scenarios typically contain multiple, sequenced processes. These processes have, of course, relationships with each other in the sense that the outcome of one process directly relates to the initial state of a subsequent process. As an example, consider a sequence of processing steps associated with Iceberg Lettuce as follows: Slicing > Performic Acid Washing > Water Washing > Storage. In this example, any contamination that has accumulated during slicing will be brought directly into the washing stages, therefore the effectiveness of the washing in killing certain microorganisms relates partly to how much of a microorganism is present after slicing. In order to model such situations, it was decided as early as Work Package about the definition of requirements that it is very desirable to link the results of processes together so that the predictions generated for one process directly become the inputs for the next process in sequence. The SOPHY software has met this objective by allowing variables to be linked together. Namely instead of a user entering a value for a given parameter, the user can choose to “carry over” the results from a previous process in a scenario and use that as an input to a model. When a user links a whole sequence of processes together, the SOPHY software is capable of calculating the effects that parameters very early on in a scenario have on the rest of the scenario right up until the end. To take the example processing steps above: If a user linked the output of a predicted log CFU/g value from the slicing stage and let that become the input log CFU/g value for the performic acid washing stage, and if the user let the output log CFU/g value of the performic acid washing stage become the input of the water washing stage, and so on, then the SOPHY software can take the chain-wide effects of that log CFU/g value into account across the whole scenario. The effect of this is very powerful, since it means that a user can experiment with optimisations at the slicing stage and directly see their effect at the final storage stage. Process linking has been successfully implemented into the SOPHY software to meet this objective.
When a scenario has had its processes linked up, the scenario can be run. Running a scenario means that the SOPHY software considers each process along the whole chain which has been defined and calls upon the models that have been selected in order to generate predictions. This activity is repeated from all processes along the chain and starts with processes in the first swimlane. When all processes in a given swimlane have had their predictions resolved (there may be many models per process), the software proceeds to the next swimlane and carries over any predicted values from processes in previously calculated swimlanes should the values be linked. The objective is to generate chain-wide predictions for a number of quality / safety indexes, and SOPHY successfully achieves this with this mechanism. At each process step, Excel models have process-specific parameters injected into them; output values are then extracted in the form of one or more scalar output values, as well as a 2D time / value array. The 2D time / value array is stored, alongside the scalar output values, in the database alongside other details about each executed process step. When all process steps have been executed, the database will be populated with a new, large set of values, which provide predictions at each step of the chain. One particular challenge is that every single Excel model starts generated predictions from a time of zero onwards. However, in a scenario, each process has a duration specified. As an example, if we had a sequence of three processes where the first process lasts for 3 minutes, the second process lasts for 60 minutes, and the third process lasts for 5 minutes, then the timings for the third process’ predictions actually start at 63 minutes. In other words the timings of all results from the individual 2D arrays extracted from models need to be stitched together in such a way that they can present a prediction accurately over the time expressed across all processes in a scenario. Stitching has been particularly challenging in the SOPHY software, yet the rewards in achieving this objective is that multiple “traces” of predictions across time can be generated in a graph.
Whilst the graphs that are generated by the SOPHY software provide quite detailed predictions of the behaviour of microorganisms over the duration of a scenario, the software can also make use of any thresholds that have been set by modelling experts. The software does so by changing the colour of each process step as it has been executed in a scenario. Should the process step generate any single prediction which falls outside of the boundaries described by a modelling expert, the software will set the process step to the colour red, which indicates a warning. In this way, a user gets a very quick signal that something may be falling outside of best practices in their scenario and can investigate further by hovering their mouse over the process step, at which point a pop-up box appears with more details about the results that have been predicted.
Furthermore, end users can also benefit from any information sheets which have been integrated by modelling experts into models. When a model is selected as being relevant to a process step (via model selection as described above), then user sees one or more hyperlinks when the scenario is run. Clicking on the hyperlink leads the user to a PDF document which contains information which must surely be relevant to the process in question, since the conditions of the process led a given model to be selected based on its relevance – since the PDF documentation attached to the model is relevant for the model, it is also relevant for the process.
In summary, the key objectives of allowing users to specify a processing chain, with each process having its conditions described, and subsequently being able to generate visual results based on the predictions generated by the models, has been achieved. Users can successfully experiment with product selections and parameters to see the effects on both quality and safety across the chain, although the scope of the products available for selection is somewhat limited to range of products supported by the uploaded models at the time of writing.
WP6
To validate the mathematical models (both existing and new developed through the project) studies were conducted using selected food products under various conditions. Microbial growth and quality changes of foods were evaluated under controlled processing and storage conditions and the experimental data were compared with the respective ones that derive from the mathematical models. For designing and implementing the pilot studies, previously developed kinetic models based on data from commercially available products were validate under dynamic temperature conditions. Furthermore, the resulted equations were tested for new food products which present similarities to existing products. The pilot studies assisted the optimisation of the mathematical models, evaluating their applicability and making suggestions for their improvement, giving input to the work package about the development of quality and safety models.
A typical polynomial, the Arrhenius and the Belehradek models were used to describe the maximum specific growth rate of pseudomonads and LAB in fresh cut salads. The developed models were validated by correlating the predictions of each model with the populations of pseudomonads or LAB that grew in each salad during storage under dynamic conditions. The polynomial and the Arrhenius model predictions showed satisfactory agreement with the observed populations of pseudomonads in romaine and iceberg lettuce. The predictions by the romaine models, showed satisfactory agreement with the observed populations of pseudomonads and LAB in different mixed leafy salads during storage at isothermal and variable conditions. These mixed products consisted from lollo verde, lollo rosso and rocket (salad 1) and iceberg, romaine and radicchio (salad 2). The rocket polynomial model gave satisfactory predictions of LAB counts in mixed salad (lollo verde-lollo rosso-rocket and iceberg-romaine-radicchio) for storage temperatures higher than 5⁰C. The iceberg model predictions showed satisfactory agreement with the observed populations of pseudomonads in MAP shredded lettuce and the mixed salad (escarole, lollo rosso, radicchio, endive). The romaine model gave satisfactory predictions in LAB levels, mainly in the mixed leafy salad (escarole, lollo rosso, radicchio, endive). The rocket model predictions showed also satisfactory agreement with the observed populations of pseudomonads in lolla rose, escarole, endive and radicchio during isothermal storage at 5 and 15⁰C. The results have shown that the behaviour of the specific spoilage microorganisms may be adequately described by the models presented in Task 4, which will be further add high value to the SOPHY software.
Regarding microbial-independent quality indices of leafy salads, the developed mathematical models on PAL activity for iceberg lettuce and the mathematical models on quinone content in iceberg and romaine lettuce were validated under variable temperature conditions. Vitamin C loss was found to be adequately described by an apparent first order reaction for iceberg lettuce, rocket and mixed leafy salads. The predicted from the model vitamin C content was well correlated to the obtained experimental values at dynamic conditions. A total mathematical model was developed that estimates the burst strength (BS) value for rocket and romaine lettuce samples at any combination of %CO2, storage temperature and time. The predicted from the model BS values were well correlated to the obtained experimental BS values. The validated models for multispectral imaging analysis will constitute and additional tool for the prediction and the description of the quality decay of leafy salads. Since these analyses are not time-consuming, the incorporation of these models to the SOPHY software will facilitate the description of the spoilage level in such products in a very easy and fast way.
The predictive models which were developed for rocket salad were also used to estimate the quality of dandelion salad. The predictions were satisfactory for LΑB growth and texture (burst strength). New kinetic models were developed for pseudomonads, texture, vitamin C, PAL activity and sensory scoring (overall quality) in dandelion salad and were validated at variable conditions.
Developing kinetic models for pre-packed salads can serve as an effective tool in the development of new products in the fresh produce food sector. This allows the use of the models introduced in the SOPHY software in the dynamic temperature conditions of the real chill chain. Additionally, the comparison of the mathematical models between similar green vegeratables supported the use of these models for new vegetable products for which specific equations are not available, and yet estimate their behaviour after different processing and storage conditions. The recently introduced fresh dandelion leaves were used as a case study and the validated models for rocket salad, which represented similarities to the new product (the closest available product), were applied. In parallel to this, new kinetic models were developed for dandelion, to allow for statistics and comparisons in quality deterioration under different packaging conditions. The validation study demonstrated the usefulness and applicability of the models developed for single leafy salads, for products (mixed leafy salads) consisting from these or similar leafy products.
Regarding deli salads, unified models were developed for LAB growth in acetic acid acidified products, including literature data and additional in-house data from similar products, as a function of the storage temperature, pH and initial concentration of the undissociated acetic acid in each product. The predictions of the unified models were compared with those of product-specific models, with temperature as the sole predictor variable. The developed models were validated under real chill chain conditions and showed very good agreement with the observed data in pepper salad and fava beans salad. The spoilage perception patterns of the different products were similar and thus, the proposed unified model may provide accurate predictions for the spoilage of a wide variety of acetic acid-acidified spreads, regardless of differences in the formulation (e.g. raw materials) and the manufacturing procedure. An independent storage test was conducted in order to collect additional data and evaluate the batch variability and the applicability of the developed kinetic models.

The applicability of the validated mathematical models for the microbial and microorganism-independent quality indices was evaluated by comparing the predicted values with the experimental. Model performance was evaluated based on RE value. For microbial growth the developed models were validated by correlating the predictions of each model with the microbial populations of each salad at the time of consumption. The prediction of the mathematical models were compared to the experimental values showing satisfactory agreement with the observed populations in most cases e.g. romaine model predictions with observed populations of pseudomonads in mixed leafy salad, and a mixed salad model showed satisfactory agreement with the observed populations of pseudomonads and LAB in the mixed product in the real cold chain. There were cases of underestimation, for example, when using the romain model to predict LAB growth in mixed salad and the rocket model was not able to predict the microbial growth in mixed salad.
Regarding vitamic C the average retention of L-ascorbic acid was expressed relatively to an initial, average value of day 0 of the experiment, where C represented the concentration of L-ascorbic acid in 100 g of raw material. In all cases, vitamin C loss was found to be adequately described by an apparent first order reaction. Arrhenius equation expressed adequately the temperature dependence of reaction rate of vitamin C, k.
Sensory evaluation was performed at the time of consumption and the scorings were compared to the respective values calculated by the predictive models for different leafy salad products and the validated kinetic models for mixed salad.
The overall sensory impression and the freshness scoring for the iceberg model and for the mixed salad model was similar. Colour and texture scorings were higher for the iceberg model. Based on the experimental and the respective calculated by the mathematical models quality level of mixed salad samples at the time of consumption, the remaining shelf life (SLR) at a reference temperature (Tref=4°C) was determined.
Available models referring to washing (water, electrolysed water, performic acid and chlorine) as a processing step of different products were compared to microbiological results of the salad samples in a field study by correlating predictions with actual observations. The washing step in a field study was carried out with peracetic acid AZT and 541 chemicals. Predictions could be made for the results obtained for the total viable counts (TVC; estimation based on mean value of all available models), Enterobacteriaceae and Pseudomonas. For two Pseudomonas models for baby spinach and washing with chlorine showed observed and predicted values in agreement in one case and predictions lower than the actual data in the second case. The predicted values Enterobacteriaceae models (E. coli and coliforms as approximation) showed three out of four models matching the observed data.

In summary, the developed models were validated at the conditions of the real cold chain. The quality level and remaining shelf life at predetermined times (i.e. the time of consumption) was estimated based on the predictive models developed in Task 4 and the values were compared to actual measured values of the quality indices. Although shelf life models were validated, indicating that they can be reliably used to estimate the quality deterioration of the target leafy salad during chilled storage, some variations between predicted and observed SLR values occurred and are mostly attributed to product (sample and batch) variability. The initial microbial load, physicochemical indices value and sensory scoring are taken into account in the calculations with the developed shelf life models.
The results of the field study show the potential for monitoring the quality of the target leafy salads in the real cold chain. Using the validated models, the SLR of RTE fresh cut salad can be estimated at any point of the chill chain if the temperature history is known. Developing and validating shelf life models can serve as an effective tool in the development of new products in the fresh produce food sector.
The developed models were validated with data obtained from a real processing line. The validation process indicated that most of the models used gave similar results with the actual data. In one occasion, a model overestimated the microbial load (fail safe) while in a second occasion a model underestimated the microbial load (fail dangerous), although available models could not be directly compared with the actual data, an approximation could be made.

Based on the models and their validation the software has been developed and presented to a number of industry members where they have been free to use the software and provide feedback in terms of industry applicability. The software has also been made available to modelling experts for some time to allow them to introduce models to the system. The experience of industry when using the software was evaluated using questionnaires and summarised and the results indicated improvements and customisations that can further be made to the software.
Academic partners with expertise in modelling entered models to the system and created scenarios in the same way as a standard user of the software would do. More than 100 models have been added successfully to the system simple scenarios were created to validate the outputs generated by models. Furthermore, the industry provided feedback in the form of a questionnaire. Part of the questionnaire was structured to gather a quantitative response, with responses being provided by industry in terms of a score between Very Poor and Very Good. Points were awarded to each possible response, with Very Poor scoring 1 point, whilst Very Good scored a maximum possible 5 points. The scoring mechanism has been used to quantify users’ responses in terms of their satisfaction with the software. The remainder of the questionnaire focussed on qualitative responses in the form of open questions. The quantitative responses from all feedback have been analysed to present an overall summary of user opinion towards the software. The qualitative responses have been analysed and grouped into various categories based on the sentiment of the response; although many qualitative responses differed in terms of content, they quite often shared the same sentiment or underlying opinion. An 77% of respondents found the user friendliness of the software to be either good or very good, 71% of respondents found the software design and layout to be either good or very good. Also, only 39% of respondents found the application range of the software to be either good or very good. At the time of testing, the software had a range of models installed which catered for scenarios covering iceberg lettuce processing in terms of slicing, performic acid washing, chlorine washing, water washing and storage, covering indices of Escherichia coli, Pseudomonads, sensory aspects, Quinone content, Vitamin C, PAL, Lactic acid bacteria, Chlorophyll and texture aspects. Some of the models installed in this class of models could also be more generically applied to other products / ingredients under a broader category of Fresh Salads. As more models become introduced to the software the satisfaction in terms of application range shifted in favour of either good or very good scores. People generally found the software to be fairly relevant, or having good relevance to their job or company.



Potential Impact:
The SOPHY project aimd to develop and validate a new generation of combined predictive and probabilistic models to be integrated into one single web-based software tool. The system allows users to store their data and results, but at the same time ensures data security. Thanks to this, SOPHY has a major impact on the establishment and implementation of new food production lines and development of minimal processed, high quality food products. During the project, deli salads and fresh cut salads were considered as test case – but the models are transferable to other ready-to-eat food commodities. The project addressed the potential of predictive microbiology and probabilistic models to estimate product safety, shelf life and quality under various conditions. It assists food producers to offer European consumers minimal processed, healthy and high quality products at an appropriate price. To increase the impact of SOPHY, the consortium has built on the strong contribution of the food industry. First reactions of food industry and their expectation towards the new software tool as well as willingness to purchase was investigated in an industry survey carried out at the beginning of the project. In addition, the project itself organised industry workshops informing food producers about the SOPHY software as well as general aspects of predictive microbiology, added-value of food quality and safety management and novel processing technologies. During these sessions reactions of industry and its meaning for the impact of the project was observed. Furthermore, an Advisory Platform consisting of food industry and experts in the disciplines addressed by the project was set up. The members of the Advisory Platform have regularly reviewed the project progress and were closely involved in the definition of the requirements. The project consortium involved four food producers having a major role in defining the requirements of the final software tool, ensuring industrial applicability and maximised impact for the food industry. Overall, the SOPHY consortium aimed to consider the expectations from the industry at all stages of the project, leading to industry relevant results and an increased impact.
Many food products are trading nowadays across countries or even continents. For example, tomatoes grown in Spain may end up on the plate of a consumer in the UK. Another example is garlic, where most of which is used in the European food industry is nowadays imported from China. These global supply chains make it necessary to consider legislation, standards and traditions in the different countries. To address the needs and expectations of food producers and consumers across Europe, SOPHY will followed a pan-European approach. The consortium was consisting of eleven beneficiaries from four different European countries. The wide regional distribution allows addressing the European perspective and ensures applicability and later exploitation of results for a wide range of food producers all over Europe. SOPHY has therefore a real added value by being conducted at European level. It combines experts from various fields – from predictive microbiology, food safety management to food processing, quality management and software development - and geographical regions – from UK in the North to Greece in the South. This European project allowed researchers and industry to combine different types of expertise and technology in a common framework and therefore increased the chances of success.
The beneficiaries were able to build on knowledge gained and results obtained from various research projects on regional and national level as well as previous and running European projects. In the majority of research projects in the area of predictive microbiology, food processing and safety/quality assurance at least one of the project beneficiaries was involved. This ensured cooperation between the initiatives and use of previous research results, simultaneously enlarging the impact of the project. The networks of former projects and activities has also contributed to building up a European contact platform being the basis for the Advisory Platform of the SOPHY project. Members of the Advisory Platform came from industry as well as research.
A wide range of dissemination and exploitation was carried out in the frame of the SOPHY project to address different target groups in an appropriate manner. Special focus was laid on communicating project results to the food industry, which are the potential customers of the newly developed software solution. The aim of the dissemination strategy towards the food industry was to contribute to the knowledge transfer among scientist and building a bridge between science and food industry. Education and training of SMEs, accomplished by eight successful workshops, was considered as a key aspect, as this directly supports food safety. These activities were also be used for the later exploitation of the SOPHY software and increasing its impact. Dissemination of project results was highly encouraged and all researchers had the freedom to write scientific publications and present their results at conferences. The driving force were the researchers themselves, who had a strong interest to increase their scientific reputation. Although the general public is not the direct customer of the newly developed software, they needed to be addressed in the dissemination strategy. After several food crisis (e.g. Listeria monocytogenes in Austrian cheese, dioxin in German eggs, EHEC in German sprouts), European consumers feel insecure. They do not know exactly, which measures industry or regulating bodies take to ensure safety of food. This lack of appropriate information paired with fear is a difficult mixture, which can ruin whole food industry sectors (e.g. UK beef producers after the BSE crisis). The SOPHY consortium contributed to the improvement of the general understanding to rebuild the shaken confidence of consumers in the European food industry. Primarily, the SOPHY film, an EURONEWS documentary, had a high impact and is available in thirteen languages.
The results of the SOPHY project improved competitiveness of all participants through knowledge generated during the project. Apart from the SOPHY software, the main exploitation perspectives are related to the “SOPHY - Data Base on Shelf Life of food under various conditions”, the “GroPIN microbial modelling Data Base”, and “Technical Information Sheets”.

List of Websites:
www.sophy-project.eu

-

ttz Bremerhaven (Germany), coordinator
Research and technology transfer centre
Department of Food Technology & Bioprocess Engineering
Website: http://www.ttz-bremerhaven.de
Contact: Christine Jewan
-
National Technical University of Athens (Greece)
University
Website: http://www.ntua.gr/index_en.html
Contact: Prof. Petros Taoukis
-
National Agricultural University of Athens (Greece)
University
Website: http://www.aua.gr
Contact: Dr. Panagiotis Skandamis
-
Campden BRI (United Kingdom)
Research organization
Website: http://www.campden.co.uk
Contact: Craig Leadley
-
University of Birmingham (United Kingdom)
University
Website: http://www.birmingham.ac.uk
Contact: Dr. Serafim Bakalis
-
ChainPoint B.V. (The Netherlands)
Company for software development
Website: http://www.chainpoint.com
-
Neue Mayo Feinkost GmbH (Germany)
Deli salads producer
Website: http://www.mayo-feinkost.de
Contact: Philip Harland, Thorsten Steinert
-
Bryans Salads LTD (United Kingdom)
Fresh cut salads producer
Website: http://www.bryanssalads.co.uk
Contact: Jon Bragg
-
Biozoon GmbH (Germany)
Food producing and research SME for innovative products
Website: http://www.biozoon.de
Contact: Dr. Alexandru Rusu
-
Heinemann GmbH (Germany)
Salad and vegetable prodcuer
Website: http://www.heinemanngmbh.de
Contact: Benjamin Heinemann
-
CE Kalamarakis-Kalas SA (Greece)
Deli salads producer
Website: http://www.kalas.gr
Contact: Dimitris Natsiopoulos

final1-sophy-logo-rz-best-quality.pdf

Powiązane dokumenty

• final1-sophy-logo-rz-best-quality.pdf