On the use of in-silico simulations to support experimental design: A case study in microbial inactivation of foods

Optimal experimental design (OED) for the growth rate of microbial populations. Are they really more "optimal" than uniform designs?

Secondary growth models from predictive microbiology can describe how the growth rate of microbial populations varies with environmental conditions. Because these models are built based on time and resource consuming experiments, model-based Optimal Experimental Design (OED) can be of interest to reduce the experimental load. In this study, we identify optimal experimental designs for two common models (full Ratkowsky and Cardinal Parameters Model (CPM)) for a different number of experiments (10-30). Calculations are also done fixing one or more model parameters, observing that this decision strongly affects the layout of the OED. Using in silico experiments, we conclude that OEDs are more informative than conventional (equidistant) designs with the same number of experiments. However, OEDs cluster the experiments near the growth limits (Xmin and Xmax) resulting in impractical designs with aggregated experimental runs ~10 times longer than conventional designs. To mitigate this, we propose a novel optimality criterion (i.e., the objective function) that accounts for the aggregated time. The novel criterion provides a reduction in parameter uncertainty with respect to the conventional design, without an increase in the experimental load. These results underline that an OED is only based on information theory (Fisher information), so the results can be impractical when actual experimental limitations are considered. The study also emphasizes that most OED schemes identify where to measure, but do not give an indication on how many experiments should be made. In this sense, numerical simulations can estimate the parameter uncertainty that would be obtained for a particular experimental design (OED or not). These results and methodologies (available in Open Code) can guide the design of future experiments for the development of secondary growth models.

The importance of what we cannot observe: Experimental limitations as a source of bias for meta-regression models in predictive microbiology

Meta-regression models have gained in popularity during the last years as a way to create more generic models for Microbial Risk Assessments that also include variability. However, as with most meta-analyses and empirical models, systematic biases in the data can result in inaccurate models. In this article, we define experimental bias as a type of selection bias due to the practical limitations of microbial inactivation experiments. Conditions with extremely high D-values (i.e. slow inactivation) need very long experimental runs to cause significant reductions. On the other hand, when the D-value is extremely low, not enough data points can be gathered before the microbial population is below the detection limit. Consequently, experimental designs favour conditions within a practical experimental range, introducing a selection bias in the D-values. We demonstrate the impact of experimental bias in meta-regression models using numerical simulations. Models fitted to data with experimental bias overestimated the z-value and underestimated variability. We propose a rapid heuristic method to identify experimental bias in datasets, and we propose truncated regression to mitigate its impact in meta-regression models. Both methods were validated using simulated data. Thereafter the procedures were tested by building a meta-regression model for actual data for the inactivation of Bacillus cereus spores. We concluded that the dataset included experimental bias, and that it would cause an overestimation of the microbial resistance at high temperatures (>120 °C) for classical meta-regression models. This effect was mitigated when the model was built using truncated regression. In conclusion, we demonstrate that experimental bias could potentially result in inaccurate models for predictive microbiology. Therefore, checking for experimental bias should be a routine step in meta-regression modelling, and be included in guidelines on data analysis for meta-regression.

Multiobjective Optimization of a Frying Process Balancing Acrylamide Formation and Quality: Solution Analysis and Uncertainty Propagation

In this study, we performed multi-objective model-based optimization of a potato-frying process balancing between acrylamide production and a quality parameter (yellowness). Solution analysis revealed that, for most of the Pareto solutions, acrylamide levels exceeded the EFSA recommendation. Almost equivalent optimal solutions were found for moderate processing conditions (low temperatures and/or processing times) and the propagation of the uncertainty of the acrylamide production model parameters led to Pareto fronts with notable differences from the one obtained using the nominal parameters, especially in the ranges of high values of acrylamide production and yellowness. These results can help to identify processing conditions to achieve the desired acrylamide/yellowness balance and design more robust processes allowing for the enhancement of flexibility when equivalent optimal solutions can be retrieved.

An experimental design tool to optimize inference precision in data-driven mathematical models of bacterial infections in vivo

The management of bacterial diseases calls for a detailed knowledge about the dynamic changes in host-bacteria interactions. Biological insights are gained by integrating experimental data with mechanistic mathematical models to infer experimentally unobservable quantities. This inter-disciplinary field would benefit from experiments with maximal information content yielding high-precision inference. Here, we present a computationally efficient tool for optimizing experimental design in terms of parameter inference in studies using isogenic-tagged strains. We study the effect of three experimental design factors: number of biological replicates, sampling timepoint selection and number of copies per tagged strain. We conduct a simulation study to establish the relationship between our optimality criterion and the size of parameter estimate confidence intervals, and showcase its application in a range of biological scenarios reflecting different dynamics patterns observed in experimental infections. We show that in low-variance systems with low killing and replication rates, predicting high-precision experimental designs is consistently achieved; higher replicate sizes and strategic timepoint selection yield more precise estimates. Finally, we address the question of resource allocation under constraints; given a fixed number of host animals and a constraint on total inoculum size per host, infections with fewer strains at higher copies per strain lead to higher-precision inference.

Guidelines for the design of (optimal) isothermal inactivation experiments

Kinetic models are nowadays a basic tool to ensure food safety. Most models used in predictive microbiology have model parameters, whose precision is crucial to provide meaningful predictions. Kinetic parameters are usually estimated based on experimental data, where the experimental design can have a great impact on the precision of the estimates. In this sense, Optimal Experiment Design (OED) applies tools from optimization and information theory to identify the most informative experiment under a set of constrains (e.g. mathematical model, number of samples, etc). In this work, we develop a methodology for the design of optimal isothermal inactivation experiments. We consider the two dimensions of the design space (time and temperature), as well as a temperature-dependent maximum duration of the experiment. Functions for its application have been included in the bioOED R package. We identify design patterns that remain optimum regardless of the number of sampling points for three inactivation models (Bigelow, Mafart and Peleg) and three model microorganisms (Escherichia coli, Salmonella Senftenberg and Bacillus coagulans). Samples at extreme temperatures and close to the maximum duration of the experiment are the most informative. Moreover, the Mafart and Peleg models require some samples at intermediate time points due to the non-linearity of the survivor curve. The impact of the reference temperature on the precision of the parameter estimates is also analysed. Based on numerical simulations we recommend fixing it to the mean of the maximum and minimum temperatures used for the experiments. The article ends with a discussion presenting guidelines for the design of isothermal inactivation experiments. They combine these optimum results based on information theory with several practical limitations related to isothermal inactivation experiments. The application of these guidelines would reduce the experimental burden required to characterize thermal inactivation.