that non-ideal experimental settings are not necessarily identified and the parameter
space has to be iteratively reduced step by step. Narrowing down the design space by
using statistical DoE requires a lot of time and experimental effort, especially in
cases where a high number of relevant factors are targeted. At the same time, the
experiments can be limited in their information content, constraining the outcome of
the optimization studies [8–10]. This generally results in a small increase in process
knowledge only.
To reduce the number of experiments and increase the process understanding
during the design and optimization of bioprocesses, a novel model-assisted design of
experiments (mDoE) concept was recently introduced [11–13]. It combines the
benefits of statistical DoE with a mathematical process model as a virtual representation of the bioprocess, called a digital twin. Although the term “digital twin” has
not yet been defined across different parts of the industry, in bioprocesses they are
intended to be a virtual counterpart of the bioprocess for the entire life cycle of the
biopharmaceutical production process. In the context of mDoE, digital twins consist
of a mathematical process model, which have gained increased importance in the last
decades. They can be applied to design [14–16], control [17–19], and optimize
[20, 21] biopharmaceutical production processes. The main intention of a mathematical model is to find solutions by analyzing the model in order to propose targeted
experiments [22]. As they contribute to a scientific understanding of the process
variables and their impact on the final product, mathematical process models in the
field of biopharmaceutical production processes are now considered to be a sustainable part of QbD [7, 12, 23, 24].
2 Design of Experiments Methods
Even if traditional trial-and-error and one-factor-at-a-time methods are still used,
advanced statistical DoE methods are applied more frequently in the field of
biopharmaceutical process development [25–27]. They can be used for the statistical
and systematic planning of experiments for hypothesis testing and/or the optimization of process variables (namely, “factors”) with regard to the desired outcome,
called “response” (e.g., product titer, product quality) [7, 28, 29]. In general, the
process development based on DoE methods leads to a certain reduction in the
number of experiments to be done in practice compared to one-factor-at-a-time
approaches. In the context of designing biopharmaceutical production processes,
they were used in the upstream as well as in the downstream part. As an example for
the design of a bioprocess, Zhang et al. (2013) implemented a screening design to
identify active parameters for the development of a serum-free medium for the
cultivation of a recombinant CHO cell line. Afterward, the process parameters
were optimized, and a fed-batch strategy was designed [30]. As an example for the
part of product purification, Horvath et al. (2010) used a screening design with eight
experiments to determine the effect of different process parameters on the isoelectric
Digital Twins and Their Role in Model-Assisted Design of Experiments
33
space has to be iteratively reduced step by step. Narrowing down the design space by
using statistical DoE requires a lot of time and experimental effort, especially in
cases where a high number of relevant factors are targeted. At the same time, the
experiments can be limited in their information content, constraining the outcome of
the optimization studies [8–10]. This generally results in a small increase in process
knowledge only.
To reduce the number of experiments and increase the process understanding
during the design and optimization of bioprocesses, a novel model-assisted design of
experiments (mDoE) concept was recently introduced [11–13]. It combines the
benefits of statistical DoE with a mathematical process model as a virtual representation of the bioprocess, called a digital twin. Although the term “digital twin” has
not yet been defined across different parts of the industry, in bioprocesses they are
intended to be a virtual counterpart of the bioprocess for the entire life cycle of the
biopharmaceutical production process. In the context of mDoE, digital twins consist
of a mathematical process model, which have gained increased importance in the last
decades. They can be applied to design [14–16], control [17–19], and optimize
[20, 21] biopharmaceutical production processes. The main intention of a mathematical model is to find solutions by analyzing the model in order to propose targeted
experiments [22]. As they contribute to a scientific understanding of the process
variables and their impact on the final product, mathematical process models in the
field of biopharmaceutical production processes are now considered to be a sustainable part of QbD [7, 12, 23, 24].
2 Design of Experiments Methods
Even if traditional trial-and-error and one-factor-at-a-time methods are still used,
advanced statistical DoE methods are applied more frequently in the field of
biopharmaceutical process development [25–27]. They can be used for the statistical
and systematic planning of experiments for hypothesis testing and/or the optimization of process variables (namely, “factors”) with regard to the desired outcome,
called “response” (e.g., product titer, product quality) [7, 28, 29]. In general, the
process development based on DoE methods leads to a certain reduction in the
number of experiments to be done in practice compared to one-factor-at-a-time
approaches. In the context of designing biopharmaceutical production processes,
they were used in the upstream as well as in the downstream part. As an example for
the design of a bioprocess, Zhang et al. (2013) implemented a screening design to
identify active parameters for the development of a serum-free medium for the
cultivation of a recombinant CHO cell line. Afterward, the process parameters
were optimized, and a fed-batch strategy was designed [30]. As an example for the
part of product purification, Horvath et al. (2010) used a screening design with eight
experiments to determine the effect of different process parameters on the isoelectric
Digital Twins and Their Role in Model-Assisted Design of Experiments
33