Modeling, monitoring, and control: a growing opportunity for vaccine production
Vaccine Insights 2026; 5(7), 389–401
DOI: 10.18609/vac.2026.049
Process control matured first in the chemical and petrochemical industries. Several of the methods now regarded as standard were developed there to improve the consistency and efficiency of large continuous plants; model predictive control is a clear example, emerging in the form of Dynamic Matrix Control from petroleum refining around 1980 and spreading through the chemical sector well before it reached the production of biologics. Biopharmaceutical manufacturing took up these ideas more slowly. The reasons were as much economic as technical. High product margins meant limited pressure to optimize, automate, or operate close to constraints: a process that worked was rarely questioned, and much of the operation stayed manual, governed by fixed recipes and operator experience. Investment flowed to drug discovery and to assure quality rather than to tighten the process itself. This situation has shifted. Margin pressure, together with a greater focus on shortening time to market, has increased the value of rapid determination of suitable operating conditions, reliable scale-up from development to manufacturing, and automation across the production line. None of these can be done well from a fixed recipe alone; each needs a usable description (some form of dynamic modeling) of how the process behaves. At the same time, the second wave of neural networks, the deep-learning era, changed how mathematical models are perceived in biopharma. Where models were once seen as optional, the abundance of data from modern instrumentation made them suddenly attractive, both to make sense of that data and as the backbone of monitoring and control. Machine learning brought mathematical modeling back into the conversation, and with it the older, well-founded methods of process systems engineering that had been waiting for exactly this combination of data and motivation. Vaccine production runs on living systems whose internal state is only partly visible from the outside, and it now has both the data and the incentive to exploit dynamic models, software sensors and model-based control. We believe that these tools have become essential for the bioindustry, and we illustrate the point with examples from our own work, from model fitting and identification, through state and rate estimation, to closed-loop control.