In this manuscript, the impact of the experimental design on the
estimation of cardinal model parameters is evaluated. In a first step,
the effect of temperature, pH and water activity is considered
separately, i.e., it is evaluated how the model describing the effect
of only one factor can be identified the most efficiently. Hereto, an
equidistant design is compared to a D-optimal (based) design (OED/
PE design) as proposed in Bernaerts et al. (2005). Based on real
experimental data and simulation studies, it was proven that the
OED/PE-inspired designs, which are based on the model’s sensitivity functions, generally yield much better results than typical
equidistant designs. By selecting the four experimental levels based
on the sensitivity functions, a more realistic description of the
behavior around optimal conditions is obtained.
In the second step, focus was on the efficient and accurate
estimation of the ten parameters of the extended cardinal model
that describes the effect of temperature, pH and water activity on
the microbial growth rate. In this step, equidistant level selection
was compared to an OED/PE-based experimental design. For both
approaches, a full factorial design and a Latin-square experimental
plan were constructed and evaluated. From the simulation case
studies presented here, it can be stated that the ten parameters of
the extended cardinal model can be equally well defined from an
equidistant design as from a D-optimal-based design. In addition,
reducing the experimental load from 64 to 12 experiments by
constructing a Latin-square design does not hamper the parameter
estimation procedure.
In a following study, we aim to evaluate how these observations
hold for real experimental data. Whereas the two case studies
shown here indicate that good parameter estimation can be expected for all designs, in reality, this might not be true. For instance,
D-optimal design might in reality yield less accurate results as the
selection of its experimental levels is highly defined by the a priori
estimation of the model parameters, which can be inaccurate.
Moreover, the definition of the experimental region considered
always emerges from the a priori knowledge, which might be
incomplete or incorrect. When preliminary information about the
studied strains is not at hand, the randomly chosen experimental
region can be an over- or underestimation of the actual growth
domain. In this case, accurate parameter identification can be
impossible due to a lack of information. Also, the reduction to 12
experiments significantly reduces the overall experimental information and renders the estimation procedure more sensitive
to (small) experimental or biological variability.
In this manuscript, the impact of the experimental design on the
estimation of cardinal model parameters is evaluated. In a first step,
the effect of temperature, pH and water activity is considered
separately, i.e., it is evaluated how the model describing the effect
of only one factor can be identified the most efficiently. Hereto, an
equidistant design is compared to a D-optimal (based) design (OED/
PE design) as proposed in Bernaerts et al. (2005). Based on real
experimental data and simulation studies, it was proven that the
OED/PE-inspired designs, which are based on the model’s sensitivity functions, generally yield much better results than typical
equidistant designs. By selecting the four experimental levels based
on the sensitivity functions, a more realistic description of the
behavior around optimal conditions is obtained.
In the second step, focus was on the efficient and accurate
estimation of the ten parameters of the extended cardinal model
that describes the effect of temperature, pH and water activity on
the microbial growth rate. In this step, equidistant level selection
was compared to an OED/PE-based experimental design. For both
approaches, a full factorial design and a Latin-square experimental
plan were constructed and evaluated. From the simulation case
studies presented here, it can be stated that the ten parameters of
the extended cardinal model can be equally well defined from an
equidistant design as from a D-optimal-based design. In addition,
reducing the experimental load from 64 to 12 experiments by
constructing a Latin-square design does not hamper the parameter
estimation procedure.
In a following study, we aim to evaluate how these observations
hold for real experimental data. Whereas the two case studies
shown here indicate that good parameter estimation can be expected for all designs, in reality, this might not be true. For instance,
D-optimal design might in reality yield less accurate results as the
selection of its experimental levels is highly defined by the a priori
estimation of the model parameters, which can be inaccurate.
Moreover, the definition of the experimental region considered
always emerges from the a priori knowledge, which might be
incomplete or incorrect. When preliminary information about the
studied strains is not at hand, the randomly chosen experimental
region can be an over- or underestimation of the actual growth
domain. In this case, accurate parameter identification can be
impossible due to a lack of information. Also, the reduction to 12
experiments significantly reduces the overall experimental information and renders the estimation procedure more sensitive
to (small) experimental or biological variability.
การแปล กรุณารอสักครู่..
