Product Purity Prediction and Anomaly Detection for an Automated Peptide Manufacturing Platform

Abstract

This thesis aims to develop and deploy a method of predicting product purity and automating anomaly detection for Mytide Therapeutics’ peptide manufacturing platform. A baseline study revealed how early purity prediction and anomaly reporting could decrease the production cycle time, manual data review, and chemical waste produced by the synthesis process. The most important tool for making purity predictions is UV absorption on the byproducts and excess reagents that come out of the reactor, where the peptides are made. A large part of this thesis was improving the quality of the UV data in order to make purity predictions using the improved UV traces. Sensor data from historical runs, including pressure, temperature, and flow rates, were analyzed to characterize several common anomalies. The reporting system takes in live data and alerts the relevant parties when the limits are reached, so that corrective action can be implemented quickly. The anomaly tracking code also generates a report to either be viewed on the user interface or stored in the backend database with the run’s historical data. Implementation of the described system improvements had several positive impacts on the workflow. The live anomaly alerts allowed for issues to be reported to the relevant parties upon occurrence, which increased the uptime of the system. The anomaly report, which is tagged to each peptide synthesis run, allows for historical data evaluation and easy decision-making for advancing the peptide to the next step of the process. The purity prediction allowed for earlier identification of certain poor-purity peptides by 27% of the production time. Together, these system improvements helped to advance the company’s peptide manufacturing platform towards total automated decision-making.