Cell therapy products are revolutionizing multiple therapeutic areas; to maintain this pace of innovation, manufacturing solutions must adapt

Bayesian cell therapy process optimization

“…Automation and closing of manufacturing processes are seen as key drivers for making cell therapies affordable and accessible to large patient populations (Abraham et al, 2021;Heathman et al, 2015).…”

“…Automation and closing of manufacturing processes are seen as key drivers for making cell therapies affordable and accessible to large patient populations (Abraham et al, 2021; Heathman et al, 2015). The required time and monetary investments are often disincentivized by high time‐to‐market pressures, costly clinical trials, and the complexity of the underlying biological mechanisms which are often not completely understood (Krishna et al, 2021; Silva et al, 2022).…”

“…The subset of inputs that are known and controllable are typically referred to as the process parameters , while the outputs of interest are referred to as the objectives . In practice, unknown and uncontrollable process inputs (e.g., due to limited knowledge, involvement of manual operations, and biological variability) make the outcome of these bioprocesses considerably variable (Abraham et al, 2021; Heathman et al, 2015). Here, we will refer to this variability as noise , and extend the process representation with an error or noise term: $$$F(x)+\varepsilon =y$$$…”

“…Cell therapies are changing the treatment paradigm in cancer. Currently, the manufacturing processes to produce these therapies mainly consist of manual and open processing steps, confining the operations to high‐grade cleanrooms in the hands of specialized operators, which prevents them from addressing large patient populations at a sustainable price (Abraham et al, 2021; Fritsche et al, 2020; Harrison et al, 2017; Heathman et al, 2015; Szkodny & Lee, 2022; Thielen et al, 2022). Closing and automating these manufacturing processes can increase their scalability, and lower manufacturing costs by up to 50% (Breig & Luti, 2021b; Emerson & Glassey, 2021; McCoy et al, 2020).…”

“…This is in stark contrast with the conditions encountered in cell therapy process optimization, as most of these therapies are autologous with a highly variable starting cell source and a labor‐intensive manufacturing process, introducing a significant amount of process noise. In addition, days or even week‐long process runtimes create the necessity for planning multiple experiments in parallel to expedite the process development trajectory (Abraham et al, 2021; Heathman et al, 2015). Fortunately, modifications to the standard BO framework have been suggested to make it compatible with parallel and noisy experimentation (Frazier, 2018).…”