Artificial intelligence is playing an increasingly important role in bioprocessing. Kai Touw explains Sartorius Stedim Biotech’s approach to these developments.
Optimising bioprocesses in the pharmaceutical industry is becoming an increasingly complex job for process operators. Batches sometimes have to be rejected if deviations cannot be resolved during the process. What if the data becomes too complex to manage? One strategy, taken by an increasing number of pharmaceutical companies, is using machine learning and other forms of artificial intelligence (AI). Machine learning is a method of data analysis that automates analytical model building. It is a branch of artificial intelligence based on the idea that systems can learn from data, identify patterns and make decisions with minimal human intervention.
Kai Touw, biopharma market manager at Sartorius Stedim Data Analytics, finds himself in a transforming market in which AI is becoming the standard, from early-stage drug discovery to prescribing treatment options. “The use of AI is growing steadily within the biopharma industry, with a projected market volume reaching $10 billion by 2024”, Touw estimates.
Sartorius Stedim Data Analytics develops algorithms that suit the specific needs of pharmaceutical companies. “Process operators want to know how material attributes and process parameters influence the quality attributes of the product and the performance indicators of the process. In other words, does our product meet the quality specifications? Are we in control of our process? And how can we optimise the process?” With more traditional methods of data analysis, it is only possible to look back at the data of previous batches and make the necessary adjustments to avoid a suboptimal process the next batch.
“The advantage of AI tools is faster processing of larger quantities of data”
“Machine Learning algorithms help predict what will happen in the next based on the current and historical data”, he explains. “The advantage of those AI tools is the faster processing of larger quantities of data. Moreover, using models for prescriptive analytics makes it possible to simulate the process based on the available data. Using a user-friendly dashboard, the operator is not only able to see the results, but also gets specific advice to adjust the process and minimise variability.” Touw gives an example of one of his first clients a couple of years ago: “A world-leading pharmaceutical company used our real-time multivariate data analysis tool for fermentation of a vaccine product. During fermentation, the tool provided an alert for a deviating process. Within 30 minutes, the deviation was linked to a faulty sensor. The operators replaced this sensor, saving the batch.”
In the light of these developments, it is easy to think of self-managing algorithms. “Intervening in a validated process by using AI is a huge step in pharma. It is theoretically possible to create such algorithms, but pharmaceutical manufacturing is highly regulated. The real challenge is to get the process validated and fail-safe. We have to ensure that our algorithms fit in this regulated environment.”
Nevertheless, Touw is convinced that these kinds of applications in bioprocessing are the future. His goal is to move there step by step, while still making user-friendly software. “I can imagine that within a couple of years, product testing is not necessary anymore. Based on the process data of a batch, the quality of a product can already be guaranteed.”