AI-ML applications in bioprocessing: ML as an enabler of real time quality prediction in continuous manufacturing of mAbs

Abstract

As continuous manufacturing of biotherapeutics gains steam, there is an increasing interest in using machine learning (ML) techniques for real time prediction of product quality and for process control. This paper focuses on application of different ML techniques for predicting critical process attributes that are pertinent to capture and polishing chromatography. Data from pH, UV, and conductivity sensors are acquired and pre-processed. For the present case study, tree-based regression techniques (decision tree and random forest) outperformed in all cases. The final model, random forest regression model, resulted in prediction errors of <5% for all predicted attributes. The results showed that random forest models exhibit optimal performance for smaller process-scale datasets with low chances of overfitting, are computationally inexpensive and do not require a graphics processing unit. The proposed approach is well suited for implementation in a continuous mAb train for real time prediction of chromatography process attributes.