Hybrid modeling for biopharmaceutical processes: advantages, opportunities, and implementation

Abstract

Process models are mathematical formulations (essentially a set of equations) that try to represent the real system/process in a digital or virtual form. These are derived either based on fundamental physical laws often combined with empirical assumptions or learned based on data. The former has been existing for several decades in chemical and process engineering while the latter has recently received a lot of attention with the emergence of several artificial intelligence/machine learning techniques. Hybrid modeling is an emerging modeling paradigm that explores the synergy between existing these two paradigms, taking advantage of the existing process knowledge (or engineering know-how) and information disseminated by the collected data. Such an approach is especially suitable for systems and industries where data generation is significantly resource intensive while at the same time fundamentally not completely deciphered such as the processes involved in the biopharmaceutical pipeline. This technology could, in fact, be the enabler to meeting the demands and goals of several initiatives such as Quality by design, Process Analytical tools, and Pharma 4.0. In addition, it can aid in different process applications throughout process development and Chemistry, Manufacturing, and Control (CMC) to make it more strategic and efficient. This article focuses on providing a step-by-step guide to the different considerations to be made to develop a reliable and applicable hybrid model. In addition, the article aims at highlighting the need for such tools in the biopharmaceutical industry and summarizes the works that advocate its implications. Subsequently, the key qualities of hybrid modeling that make it a key enabler in the biopharmaceutical industry are elaborated with reference to the literature demonstrating such qualities.