Design of smart liquid-liquid extraction columns for downstream separations of biopharmaceuticals using deep Q-learning algorithm

We propose smart liquid-liquid extraction columns of biopharmaceuticals using deep Q-learning algorithm. In this contribution, we demonstrated the application of the tool for design of liquid-liquid extraction process for concentration of API from fermentation broth. To this end, we present the following;1) development of property model to describe solubility of API in different solvents using the nonrandom two-liquid segment activity coefficient model, 2) design the liquid-liquid extraction process for different solvent candidates commonly used in pharma industries, 3) application of deep Q-learning algorithm to optimize liquid-liquid extraction control, and 4) perform sensitivity analysis to study effect of feed fraction of API on the performance. We have validated the developed property process modelling by comparing the existing experimental data and the characteristics of diverse solvents and using sensitivity analysis. We expect that the results from this study would contribute to further development the general framework of downstream separation for the future by extending to more downstream separation processes.