Hybrid modeling reduces experimental effort to predict performance of serial and parallel single-pass tangential flow filtration

Abstract Single-pass tangential flow filtration (SPTFF) is a promising and increasingly important tool for continuous biomanufacturing and intensified bioprocessing. SPTFF operates at high product concentrations often limiting the number of process development experiments and therefore process understanding. We developed hybrid model structures to predict the concentration performance of SPTFF in serial and parallel mode with up to three membranes at various pressures, feed flows and protein concentrations. A single experiment to gather the training data for the model was performed in batch mode, significantly reducing the required amount of protein to set up the ideal process design. The hybrid model's excellent interpolation properties further reduce the duration of the training experiments. A set of mass balances controls the model variation of the two SPTFF modes as part of the hybrid model requiring only one training set for both modes. The hybrid model can be used as a Digital Twin to perform in-silico process simulations to gain in-depth process knowledge and enable model-based process control. The Digital Twin facilitates faster process transfer from batch to continuous and gives predictions on how different SPTFF modes perform under changing process conditions. This renders it a valuable tool for efficient process development and process transfer.