Multiple Intelligent Agents for Manufacturing Intensification (MIAMI): A Platform for Ranking Clonal Variation in Upstream Bioprocess Development

摘要: Antibody-based therapeutics are an important class of biotherapeutics for therapeutic applications. With the rising demand and increase in biotherapeutic products on the market, there lies the need for rapid bioprocess development. Clone selection is a critical and time-consuming step in upstream bioprocess development and it is a critical step to execute accurately. A Multiple Intelligent Agents for Manufacturing Intensification (MIAMI) is proposed to process raw data and evaluate clones of three commonly used host cells, Chinese Hamster Ovary (CHO), Escherichia coli (E. coli), and Pichia pastoris (P. pastoris). A search conducted for an IP-free protein sequence yielded the Anti-hepatitis B antibody. The whole antibody sequence was truncated to create a Fab' fragment. Gene designs for three commonly used host cells, CHO, E. coli, and P. pastoris were created using the IP-free Anti-hepatitis B Fab' fragment. The development of MIAMI identifies and addresses the necessity of creating a sophisticated code that evaluates clonal ranking based upon data sets. These data sets were collected using the IP-free Anti-hepatitis B gene designs and an existing AV4 gene design. The AV4 gene design was transformed into P. pastoris and repurposed as an inverse methanol detector. In 50mL shake flask culture, green fluorescence protein was detected when cultivating the AV4 strain using glycerol and sorbitol carbon source, while protein transcription was inhibited when using a methanol carbon source. Data collected from cultivating the AV4 strain in 800μL microtiter plates was used to develop the MIAMI software. The Anti-hepatitis B gene designs were established and characterized in 50mL shake flasks for E. coli and P. pastoris and a preliminary attempt to establish the gene design CHO. Using the data collected from automated cultivation of 8 different clones of Anti-hepatitis B E. coli and P. pastoris strains in 800μL microtiter plates scale using the TECAN, a manual ranking of the clones was performed. Scaling the cultivation up to 200mL DASGIPs microbioreactors, clonal ranking for both strains remained unchanged. A code was written in python for the processing of raw data. This was demonstrated on the collected HPLC data sets for the Anti-hepatitis B E. coli and P. pastoris strains, and the flow cytometer data set for the AV4 strain. Multiple agents were created for the development of MIAMI. An assay agent was created for analysing raw HPLC and flow cytometry data to identify and remove unwanted clonal variations. A scanning algorithm calculated the mean and standard deviation of the yields at three consecutive time points to identify a period of stable yield. A ranking algorithm takes into consideration the maximum stable yield achieved and the variability in the data point, giving these two factors a 75% and 25% weighting, MIAMI identifies the best performing clone. The MIAMI ranking came to the same conclusion as manual human ranking. The effectiveness of MIAMI was validated on the Anti-hepatitis B E. coli strain, being able to correctly identify a top performing clone with an optimal induction time, with a conservative estimate of 87% decrease in time taken when compared to manual evaluation. The MIAMI software significantly improved the timeliness of bioprocess development by accurately screening and evaluating clones. This frees up the time of the user while removing potential sources of human error. With the incorporation of further bioprocesses, MIAMI will become a powerful and effective tool for bioprocess development.