highly automated systems and closely interconnected devices, concepts like Internet
of Things (IoT) and Digital Twin come to mind. Beyond the hype around Digital
Twins, its history and modern definition are closely related to High-Throughput
Bioprocess Development. The term Digital Twin emerged in the field of Product
Lifecycle Management to increase the efficiency in product and process development [6]. Mathematical models, used for process monitoring (observers) in feedback
loops, for approximate optimal control applications (MPC), or even in real-time
optimization have existed for quite some time now [7]. Yet, the extension of these
methods including IoT, big data, and fully autonomous systems might require a new
terminology [8].
The Digital Twin, envisioned as a mirror image (an exact copy) of the physical
system that follows the complete lifecycle of the product from idea to manufacturing, is possible only if (1) an exact representation of the system in mathematical
equations is at hand and (2) the current state of all relevant elements of the real
system can be fully monitored through real-time data. In bioprocess development,
building a Digital Twin implies joining High Throughput, Omics, PAT, Machine
Learning, Bioprocess Automation, and Bioprocess Systems Engineering tools to
enable the development and operation of a biomanufacturing plant with a perfect
copy of all units from the molecular/intracellular level up to large-scale dynamics.
Such a Digital Twin is clearly far beyond the capabilities of current technologies.
Still, it defines a clear roadmap that shows the relevance of the integration of
different fields and tools to maximize the efficiency of bioprocess development.
Mathematical models, which form the basis of digital twins, support all fields of
biotechnology and bioprocess engineering [9]. This includes biochemical systems
[10], systems biology [11], metabolic engineering [12], flux balance analysis [13],
synthetic biology [14], and bioinformatics [15]. A good overview of applications of
mathematical models, as well as a proof of their slow advance in bioprocess
engineering is given by Jay Bailey [16]. Nevertheless, the complexity of biological
systems poses difficult challenges to the direct use of advanced mathematical
techniques in bioprocess development [17].
The complexity of the underlying metabolic and physiological phenomena
demands large nonlinear equation systems with a large number of unknown and
often time-variant parameters. The existing methods are too complex and computationally expensive for application in biotechnology [18–20]. Compared to general
applications in engineering [21–24], biotechnological applications typically lack
sufficient data, as well as process understanding [25–27].
Finally, the advances in artificial intelligence, especially in data-driven learning
tools, offer incredible possibilities, but need to be adapted to the specific needs of
bioprocesses, which have peculiarities (e.g., evolution of the biological system [28]),
broad population distributions, very complex chemical composition and complicated
(metabolic) reaction networks that are not present in mechanical or chemical processes [12]. Nevertheless, such mathematical tools have greatly contributed to our
understanding of the interactions between the organism and the constraints of growth
in bioreactors, as well as the elucidation of otherwise obscure intracellular
processes [13].
4
P. Neubauer et al.
of Things (IoT) and Digital Twin come to mind. Beyond the hype around Digital
Twins, its history and modern definition are closely related to High-Throughput
Bioprocess Development. The term Digital Twin emerged in the field of Product
Lifecycle Management to increase the efficiency in product and process development [6]. Mathematical models, used for process monitoring (observers) in feedback
loops, for approximate optimal control applications (MPC), or even in real-time
optimization have existed for quite some time now [7]. Yet, the extension of these
methods including IoT, big data, and fully autonomous systems might require a new
terminology [8].
The Digital Twin, envisioned as a mirror image (an exact copy) of the physical
system that follows the complete lifecycle of the product from idea to manufacturing, is possible only if (1) an exact representation of the system in mathematical
equations is at hand and (2) the current state of all relevant elements of the real
system can be fully monitored through real-time data. In bioprocess development,
building a Digital Twin implies joining High Throughput, Omics, PAT, Machine
Learning, Bioprocess Automation, and Bioprocess Systems Engineering tools to
enable the development and operation of a biomanufacturing plant with a perfect
copy of all units from the molecular/intracellular level up to large-scale dynamics.
Such a Digital Twin is clearly far beyond the capabilities of current technologies.
Still, it defines a clear roadmap that shows the relevance of the integration of
different fields and tools to maximize the efficiency of bioprocess development.
Mathematical models, which form the basis of digital twins, support all fields of
biotechnology and bioprocess engineering [9]. This includes biochemical systems
[10], systems biology [11], metabolic engineering [12], flux balance analysis [13],
synthetic biology [14], and bioinformatics [15]. A good overview of applications of
mathematical models, as well as a proof of their slow advance in bioprocess
engineering is given by Jay Bailey [16]. Nevertheless, the complexity of biological
systems poses difficult challenges to the direct use of advanced mathematical
techniques in bioprocess development [17].
The complexity of the underlying metabolic and physiological phenomena
demands large nonlinear equation systems with a large number of unknown and
often time-variant parameters. The existing methods are too complex and computationally expensive for application in biotechnology [18–20]. Compared to general
applications in engineering [21–24], biotechnological applications typically lack
sufficient data, as well as process understanding [25–27].
Finally, the advances in artificial intelligence, especially in data-driven learning
tools, offer incredible possibilities, but need to be adapted to the specific needs of
bioprocesses, which have peculiarities (e.g., evolution of the biological system [28]),
broad population distributions, very complex chemical composition and complicated
(metabolic) reaction networks that are not present in mechanical or chemical processes [12]. Nevertheless, such mathematical tools have greatly contributed to our
understanding of the interactions between the organism and the constraints of growth
in bioreactors, as well as the elucidation of otherwise obscure intracellular
processes [13].
4
P. Neubauer et al.