On-the-fly anomaly detection in autonomous HPLC experimentation at CMU Cloud Lab

The shift towards AI-driven experimental design methodologies has significantly impacted scientific research, particularly in the domain of commonly used experimental techniques with a high-dimensional set of conditions, such as High-Performance Liquid Chromatography (HPLC). Historically, the proper setup and execution of HPLC experiments relied on domain-knowledgeable human experts. The emergence of cloud laboratory environments and the automation of experimental protocols have greatly expanded the historical data availability for HPLC method development. This has brought to light the essential requirement for systems capable of validating experimental data on-the-fly, without human expert intervention. Among many factors that can adversely affect the quality and validity of HPLC, air bubbles in the system are one of the most commonly observed in a massive parallel experimental execution. Despite the rareness of such events as well as relative simplicity for individual expert analysis, their detection in an automatic matter substantially improves overall data quality for the autonomous execution of massive parallel HPLC runs. Herein, we developed an automated ML-based approach to characterize HPLC data retrospectively and autonomously validate the HPLC execution on-the-fly at the CMU Cloud Lab facility.