Artificial intelligence in pharmaceutical manufacturing: Applications, case studies, and GxP implementation...

Artificial intelligence (AI) and machine learning (ML) applications are starting to be employed within the pharmaceutical manufacturing industry to create better process controls and real-time process monitoring and inspection systems for both small-molecule and biopharmaceutical production facilities. The combination of process analytical technology (PAT) and quality by design (QbD) principles enables AI/ML to create soft sensors that perform inferential measurements, multivariate monitoring, anomaly detection, advanced control systems and computer vision inspection. A thorough understanding of Good Manufacturing Practice (GMP) requirements combined with practical predictive performance are the foundations for successful implementation within regulated environments. The paper integrates peer reviewed studies with applicable regulatory documents to create a mapping system that shows common AI use-case archetypes (monitoring, decision support, control/optimization, and inspection) while defining data requirements along with intended use and validation needs. The eight case studies demonstrate actual use cases for continuous manufacturing which includes multivariate statistical process control (MSPC), potency soft sensors, hybrid NIR-soft-sensor in-process control (IPC), interpretable data-driven model predictive control (MPC), downstream bioprocessing which includes chromatography anomaly detection & soft-sensing architectures, upstream Chinese Hamster Ovary (CHO) cultivation optimization which uses neural networks and ML-assisted automated visual inspection of sterile injectables. The examples demonstrate that in order to deploy sustainable AI/ML based applications, organizations have to create data monitoring systems that provide thorough tracking with complete traceability. Regulatory considerations throughout the lifecycle of these models are discussed that can be integrated into the pharmaceutical quality system (PQS).