Validating drug production processes need not be a headache, according to AI researchers who say machine learning (ML) could be a single answer to biopharma’s multivariate problem. The FDA defines process validation as consisting of three parts: process design (PD), process qualification (PQ), and continued process verification (CPV). The first two stages are discrete—once approved, they are complete.
CPV, in contrast, is an ongoing process that requires drug makers to track and analyze complex data for as long as the process runs. And this is a challenge, says Toni Manzano, co-founder and CSO at Aizon, an industrial artificial intelligence developer.
“CPV is in essence a continuous process which ensures that production is under control and that the critical process parameters and relevant factors are monitored and evaluated in real-time in order to ensure the expected quality,” he explains. “CPV must be a multivariate process where all the critical factors are controlled, considering their interactions as well and not individually. Nowadays, CPV, even in the best of cases, is usually limited to a single-variable approach for each factor which is time consuming and inefficient.”
To try and overcome this, Manzano and colleagues set out to see if artificial intelligence—and more specifically machine learning, a subset of the approach in which an algorithm determines “rules” based on its own analysis of data—could do the job more effectively.
In their recent study, the researchers looked at the production of a recombinant protein called candida rugosa lipase 1 (Crl1) by the yeast species Pichia pastoris under hypoxic conditions in fed-batch cultures.
Two AI models
The team used two AI models—an isolation model to detect anomalies during the batch phase of the process and a random forest model to predict required operator control actions during the semi-automated fed-batch phase.
The models outperformed traditional single-variable approaches, took a fraction of the time and, according to the authors, illustrate the potential benefits of AI in process analysis.
“The work presented here constitutes a proof-of-concept that multivariate analytics methods, based on machine learning, can be a valuable tool for real-time monitoring and control of biopharma manufacturing bioprocesses to improve its efficiency and to assure product quality,” they wrote.
For Manzano, it is this ability to find otherwise undetectable patterns in complex data with minimal operator intervention that makes AI systems an ideal fit for manufacturing operations like CPV and process control.
“AI/ML is by default a set of multivariate techniques that replicates human cognitive abilities. As in other statistical disciplines, AI/ML requires good data to create models representing the reality provided by the raw data under a multivariate approach,” he adds.
