point of a therapeutic antibody expressed in CHO cell culture. The pH, temperature,
and the time of the temperature shift were significant. These factors were evaluated
in three levels in a concluding response surface design to optimize the isoelectric
point [31].
Statistical DoE methods are solely based on user-defined selections of the experimental design and the definition of factor limits, including the definition of experimental variables and their evaluated levels [8, 32, 33]. This can lead to error-prone
decisions, iterative re-adjustments of the experimental space with several rounds of
costly and time-intensive experiments, and even to a design that simply cannot be
implemented [7]. Expert knowledge is required to select suitable boundary values
for process development and optimization using DoE [7, 34–36]. Therefore, the
combination of digital twins with DoE in mDoE offers a novel tool for the
knowledge-driven development of bioprocesses.
2.1 Screening Designs
Screening designs are intended to identify the significantly influencing factors from a
list of many potential factors [33, 37]. Therefore, different experimental designs can
be used. The most commonly used designs, called full factorial, factorial fractional,
as well as Plackett-Burman designs, are discussed.
2.1.1 Full Factorial Designs
A full factorial design can be used to examine the main effects and interactions of
one or more factors on the respective response. The design consists of two or more
factor levels and k-factors, resulting in at least a 2
k -design [38, 39]. Exemplary, the
full factorial design for three factors is given by a 2
3 -design plan, shown in Fig. 1a.
2.1.2 Reduced Full Factorial Designs
In order to reduce time-consuming and costly experiments in the case of a large
number of factors, incomplete designs, like fractional factorial and Plackett-Burman
designs, can be chosen. The fractional factorial designs, representing a reduced form
of the two-level factorial design, are based on the assumption that higher-value
interactions are irrelevant. This results in a 2
k-n -design, whereby the 2
k -design is
reduced by n steps [38, 40, 41]. A reduced form of the previously mentioned 2
3 -
design plan, a fractional factorial 2
3-1 -design, is shown in Fig. 1b. Plackett-Burman
designs, a special form of the two-level fractional factorial designs, are suitable if the
34
K. B. Kuchemüller et al.
and the time of the temperature shift were significant. These factors were evaluated
in three levels in a concluding response surface design to optimize the isoelectric
point [31].
Statistical DoE methods are solely based on user-defined selections of the experimental design and the definition of factor limits, including the definition of experimental variables and their evaluated levels [8, 32, 33]. This can lead to error-prone
decisions, iterative re-adjustments of the experimental space with several rounds of
costly and time-intensive experiments, and even to a design that simply cannot be
implemented [7]. Expert knowledge is required to select suitable boundary values
for process development and optimization using DoE [7, 34–36]. Therefore, the
combination of digital twins with DoE in mDoE offers a novel tool for the
knowledge-driven development of bioprocesses.
2.1 Screening Designs
Screening designs are intended to identify the significantly influencing factors from a
list of many potential factors [33, 37]. Therefore, different experimental designs can
be used. The most commonly used designs, called full factorial, factorial fractional,
as well as Plackett-Burman designs, are discussed.
2.1.1 Full Factorial Designs
A full factorial design can be used to examine the main effects and interactions of
one or more factors on the respective response. The design consists of two or more
factor levels and k-factors, resulting in at least a 2
k -design [38, 39]. Exemplary, the
full factorial design for three factors is given by a 2
3 -design plan, shown in Fig. 1a.
2.1.2 Reduced Full Factorial Designs
In order to reduce time-consuming and costly experiments in the case of a large
number of factors, incomplete designs, like fractional factorial and Plackett-Burman
designs, can be chosen. The fractional factorial designs, representing a reduced form
of the two-level factorial design, are based on the assumption that higher-value
interactions are irrelevant. This results in a 2
k-n -design, whereby the 2
k -design is
reduced by n steps [38, 40, 41]. A reduced form of the previously mentioned 2
3 -
design plan, a fractional factorial 2
3-1 -design, is shown in Fig. 1b. Plackett-Burman
designs, a special form of the two-level fractional factorial designs, are suitable if the
34
K. B. Kuchemüller et al.