A Hybrid Kinetic Machine Learning Model for Accelerating Cell Line Selection

The use of hybrid models, combing mechanistic and machine learning (ML), has emerged as a promising approach, contributing to the development of Industry 4.0. This work presents a hybrid model that forecasts minibioreactor (MBR) production runs of mammalian cell culture recombinant for monoclonal antibodies (mAbs), using micro-scale and small-scale cultivation data as inputs. The hybrid model consists of a ML ‘unit’ and a kinetic model; it takes single time-point measurements of final mAb titre and cell growth from the Beacon™, 24-well plates, 6-well plates, and T25 flasks of 140 Chinese hamster ovary (CHO) cell lines as inputs, to predict the metabolic profiles of biomass, mAb, glucose, glutamine, glutamate, lactate, ammonium, and the growth rate profiles for each cell line cultivation. The hybrid model is particularly accurate in forecasting cell growth and mAb profiles in the MBR with R2=0.80 and 0.88 respectively. It is reasonably accurate for ammonium (R² = 0.67, pRMSE = 10.0%), while glucose, glutamine, and lactate showed higher errors (pRMSE = 27.8%, 20.3%, 22.2%). This is due to regression inaccuracies of the kinetic part of the hybrid model, which could be improved by experimentally determining cell line-specific kinetics. This tool advances early cell line development (CLD) by enabling earlier and more accurate cell line selection, and comparison of different cell lines. It addresses the challenge of poor correlation between small-scale models and bioreactor performance, enhancing process prediction and optimisation, and reducing time and costs in biopharmaceutical bioprocess development.