Assisted Active Learning for Model-Based Method Development in Liquid Chromatography.

Area of Science:

• Analytical Chemistry
• Computational Chemistry

Background:

• Growing interest in automated methods for complex optimization problems.
• Active learning (AL) and assisted active learning (AAL) enable autonomous selection of optimal experimental conditions.
• AAL is valuable when experimentation is costly or time-consuming.

Purpose of the Study:

• To explore the application of AAL in model-based method development (MD) for liquid chromatography (LC).
• To generate initial retention models using Bayesian statistics incorporating historical data and analyte information.
• To iteratively update models and select informative experiments for efficient separation optimization.

Main Methods:

• Utilized Bayesian statistics for initial retention model generation.
• Incorporated historical data and analyte information.
• Employed an active data selection method for choosing subsequent experiments.
• Iteratively updated model parameters based on new experimental data.

Main Results:

• Demonstrated AAL effectiveness in two practical LC method development examples.
• Achieved optimized separations in a limited number of experiments by optimizing gradient slope.
• Showcased AAL's ability to leverage past knowledge and compound information.

Conclusions:

• AAL offers a flexible alternative to fixed design methods for LC separation optimization.
• AAL improves accuracy and reduces experimental runs by intelligently selecting experiments.
• The approach balances model exploitation and experimental exploration for efficient optimization.