Neural Network to Understand Process Capability and Process Intermediates Acceptance Criteria in Monoclonal Antibody Production Process

Abstract

In today’s competitive environment, it is crucial for biopharmaceutical companies to have a robust R&D pipeline and reliable manufacturing processes. To ensure success in drug manufacturing, regulatory agencies often mandate appropriate acceptance criteria for process intermediates to increase the likelihood of the drugs meeting the final release specification. When setting acceptance criteria for process intermediates, it is important to first understand process capability, or the impurity clearance of each process step. However, this process involves either challenging experimentation or an estimation method that might not be comprehensive. In this study, we propose the use of neural network to understand process capability. This approach not only will be able to delineate the relationship between the feed and product impurity level for a specific step but will also be able to define the acceptance criteria for the feed (or product from previous process step) impurity level based on a predetermined product impurity level. These acceptance criteria will enable us to determine whether or not to forward process the step based on the feed impurity level. Process impurity data are a combination of actual data collected from actual manufacturing lots and simulated data. Impurity clearance for a specific step is estimated using a conventional method and a neural network, a method that we propose in this study. Since impurity clearance is usually dependent on the input impurity level, using neural network to estimate process capability and ultimately to define process intermediates acceptance criteria has been shown to be more useful than the conventional method.