Surrogate optimization with multivariate adaptive regression splines for supercritical fluid extraction-supercritical...

When analyzing complex mixtures, on-line extraction techniques improve reproducibility and robustness in comparison to bench extractions, but they often face challenges due to the complexity of method development processes. In this study, we introduce a new optimization strategy for on-line extraction that uses a multivariate adaptive regression splines (MARS)-based surrogate optimization approach to optimize five key parameters associated with supercritical fluid extraction: flow rate, modifier concentration, back-pressure, static time, and dynamic time. A commercial on-line supercritical fluid extraction - supercritical fluid chromatography - mass spectrometry instrument was used. A mixture of 12 pharmaceuticals were selected; structural diversity metrics were encoded and used for subset selection and to evaluate molecular similarity. Hydrocodone was chosen arbitrarily as the primary analyte for parameter optimization and the primary reference for molecular similarity. The performance of the 11 other compounds were monitored. A ratio of peak area to full-width at half maximum (A/W) was the primary performance metric, to reflect both extraction efficiency and chromatographic efficiency through the optimization process. The MARS surrogate optimization algorithm identified optimal parameters within 20 runs, greatly reducing experimental effort and achieving a relative standard deviation of 3 %. An additional composite performance metric was devised to include peak symmetry and reproducibility, which further lowered variability to 2 % and reached a best-known solution in just 16 runs. Compared to traditional methods like central composite design, surrogate optimization converges faster and explores the parameter space more effectively. Surrogate optimization using MARS provides a new and efficient framework for optimizing complex extraction processes.