Consistent value creation from bioprocess data with customized algorithms: opportunities beyond multivariate analysis

Abstract

Biopharmaceutical drugs generate yearly revenues exceeding €200 billion, representing about 20% of the pharma market and its largest growing sector. However,their manufacturing requires a complex and regulated journey to transfer a patenteddrug from lab scale (from 1 ml) to production scale (up to 50 000 l). During thedevelopment of the underlying manufacturing process, a great number of potentialparameters must be considered starting from the selection of the productionorganism through the operating parameters of the bioreactor to the subsequentpurification steps. Due to the high capital expenditure until market entry andincreasing competition, biopharma companies are facing pressure for cheaperprocess development, faster time to market (13 years on average spent of 20 yearspatent lifetime, where about 18 months are dedicated to process development),and more consistent production, i.e. reducing failure runs (currently up to 5%resulting in the tens of millions euro in revenue losses per failed batch). There areindications that one-day delay in market entry could cost €0.5 million in net presentvalue per drug. In Europe, more than €3 billion are yearly spent on processdevelopment, and more than €66 billion in drug manufacturing . The challengeof fast process development and robust operation is partially due to a low degree ofdata digitalization, standardization, and central storage and a level of automatedand adaptive operation procedures including many sources for human errors [4]. Inrecent years, top management of large pharma companies has therefore attributeddigitalization as one of the main priorities.To tackle the various bottlenecks, the biopharmaceutical industry is lookingfor digital solutions within their process intensification initiatives toward processanalytical technology (PAT), continuous bioprocessing, and robotic experimentalsystems as a means of resources, capacity, cost, and risk reduction. We havean aspiring vision of the future where smart data analytics and model-based process digitalization and automation are at the heart of operational excellence inbiopharma.However, the current state-of-the-art in silico tools are often not sufficient tosupport the in vitro requirements of process intensification. In this chapter,we present advanced modeling concepts that significantly outperform the currentindustrial benchmarks for different applications, namely, process variable forecasting, product quality prediction, monitoring and control, scale-up and processoptimization.With the proposed methods, combining bioengineering know-how and customized machine learning techniques, the goal is to consistently support decisionmaking across the complicated process development and tech transfer activities,to enable for the corresponding professionals to consecutively advance toward thestandards of industry 4.0.