Better precision could improve many aspects of bioprocessing. Although optimizing a bioprocess and monitoring its progress would benefit from better precision, accomplishing those goals challenges scientists because so many factors must be considered.
In a recent review in Bioresource Technology, published by Elsevier, Rohan Jain, PhD, head of metallophores at the Helmholtz-Zentrum Dresden-Rossendorf in Dresden, Germany, and his colleagues noted: “A variety of factors like technical understanding of involved bio-kinetics, time-delayed and scale-dependent bioprocess responses, decelerating process drifts, instrumentation, process disturbances, and environmental disturbances may influence process outcome.”
These days, when a problem includes many parts and interconnected datasets, many scientists turn to tools that involve artificial intelligence (AI). In particular, Jain’s team explored the potential of machine learning (ML) in bioprocessing.
Though AI terms get thrown around a lot—and often mixed up—let’s take a look at what IBM says about ML’s place in AI: “Machine learning is a subset of artificial intelligence that allows for optimization.” Of particular importance for Jain’s work, IBM added: “When set up correctly, [ML] helps you make predictions that minimize the errors that arise from merely guessing.”
ML can be applied to many aspects of bioprocessing, including “controlling and monitoring the bioreactors” or “identifying spectroscopic errors during chromatography analysis,” Jain and his co-authors noted. Moreover, many ML-based tools can be used. For example, Jain’s team pointed out that, “Partial Lease Square (PLS) and Principal Component Analysis (PCA) methods are applicable in bioreactors.” Some ML-based tools can even be used to “quantify real-time control processes of bioreactors at pilot and commercial scale,” Jain and his colleagues explained.
Despite the potential benefits of applying ML to bioprocessing, one tool is not always as good as another. According to Jain’s team, “The real challenge for real-time application of ML in the bioprocess industry is selection of [the] right algorithm and model building as each ML-model has limitations and prediction capability varies with [the] output goal.” Plus, one tool is also not necessarily enough. In one example, Jain and his colleagues mentioned that combining an artificial neural network—a subfamily of ML—and PLS, plus an online fluorescence sensor, “forecast the real-time data for manufacturing a recombinant biopharmaceutical protein in Escherichia coli.”
Although machine learning can reduce the guesswork in bioprocessing, scientists shouldn’t guess in trying to select the right AI-based tools. For bioprocessors to make the most of this branch of AI, it takes the proper ML-based tool, probably other analytical methods, and correct data.
