Bioprocessing 4.0 emerges from a largely computation-based foundation, or at least it should. Nonetheless, lots of work remains to be done, and not necessarily in the places that even experts expect.
In writing about the biopharmaceutical industry, Jens Smiatek, PhD—an expert in computation and developing biologicals based at the Institute for Computational Physics at the University of Stuttgart—and his colleagues noted: “Some of the most important new technologies are various digital approaches, which have found their way into automatized laboratories, artificial intelligence for advanced data analysis, and computational models to study molecular or process-relevant behavior.”
Nevertheless, these scientists pointed out that current computational methods do not always consider the final product as much as they should.
When asked to discuss the most important computational method for bioprocessing, Smiatek says, “Most people think that accurate unit operation models for upstream and downstream are most important with regard to their huge impact on the bioprocess.” As Smiatek points out, though, lots of accurate and reliable computational methods exist for upstream and downstream processes. “Over the last years, mechanistic models for downstream received the largest attention,” he says. “The results are quite accurate and helped a lot to improve chromatography steps.”
On the upstream side, he also points out a few beneficial methods, including computational fluid dynamics for optimizing the design of a bioreactor or hybrid mechanistic/neural network models to predict the temporal behavior of metabolite concentration. So, which method matters the most depends on who you ask.
When you ask Smiatek, the answer is neither single and non-connected upstream nor computational methods. Instead, he sees two other challenges. The first ones are models of the operation of individual units. Smiatek says that these “are helpful but provide only limited insights into the bioprocess as a whole.” As an example, if an upstream fermentation process fails—producing little of the desired product—there is nothing to capture downstream. “Thus, it is of utmost importance to combine the individual unit–operation models into a holistic process model,” he says.
Importance of final drug product
The second crucial challenge that Smiatek mentions is: “People often ignore the importance of the final drug product, meaning the API in a suitable buffer formulation.” In particular, he says that computation must consider how the API behaves in solution. To further explain this, he notes: “As a simple example, even if one optimizes the bioprocess in terms of incredible titers, they are worth nothing if one is not able to stabilize the API in solution.”
Proteins might aggregate or partially unfold in solution, thereby impacting their function. So, scientists need “a detailed molecular understanding of the interactions in solution,” he explains. “Such an understanding can be partially provided with limitations by molecular models as well as the detailed use of molecular theories of solution.”
Most models in the past ignored this part of bioprocessing. Nonetheless, Smiatek says that promising models and theoretical insights exist from academic research. “It is clear that such models do not provide a full understanding due to the complexity of the problem, but they may help to elucidate the main principles, which can be eventually used to reduce trouble-shooting,” he explains.
In thinking about why such simulations remain largely unused in bioprocessing, Smiatek says, “Molecular approaches are time-consuming and challenging and the interpretation of the outcomes requires expert knowledge in physics and chemistry.” Instead, he regularly sees the application of inadequate or simplified differential equations, but that cannot “describe protein aggregation or specific ion effects, which explains why this field was largely ignored in the past,” Smiatek explains, being careful to note that this is just his opinion.
“I would say accurate molecular models in addition to deeper molecular insights or drug formulations would have the strongest impact in the next years to reduce trouble-shooting events as well as to optimize the shelf lives of drug products,” he adds.
To accomplish this, Smiatek explains that experts must “identify the most accurate and verified unit operation models, such that a digital bioprocess replica with suitable transformation functions can be developed.”
To really get Bioprocessing 4.0 to new levels of efficiency and accuracy, the most advanced computation should be applied. In some cases, existing methods could drive big improvements, if bioprocessors would use the techniques.