Artificial Neural Networks: An Innovative Approach Used for Elucidation of Ionization Processes in Supercritical Fluid Chromatography-Mass Spectrometry

Abstract

Understanding and predicting mass spectrometry responses in supercritical fluid chromatography-mass spectrometry (SFC-MS) is critical for optimizing detection across diverse analytes and solvent compositions. We present a novel approach using artificial neural networks (ANN) to explore the complex relationships between molecular descriptors of analytes and MS responses in different makeup solvent compositions enabling SFC-MS coupling. 226 molecular descriptors were evaluated for compounds under standardized SFC conditions, with 24 makeup solvent compositions. These makeup solvents included pure alcohols and methanol with varying concentrations of volatile additives. Our results highlight distinct ionization processes for the two most commonly used soft ionization techniques: (i) electrospray ionization (ESI), primarily involving proton or cation transfer, and (ii) atmospheric pressure chemical ionization (APCI), associated with charged ion transfer. Principal component analysis of weights assigned to molecular descriptors reveals that, in positive detection mode, these descriptors effectively differentiate ionization efficiency between ESI and APCI. In contrast, this differentiation is less pronounced in negative mode, where the variance explained is more homogeneously distributed, with stronger discrimination observed when NH₃ is used as an additive to the organic modifier. These findings provide critical insights into the influence of molecular descriptors and solvent composition on ionization efficiency, serving as a foundation for future investigations into SFC-MS optimization. This proof-of-concept underscores the feasibility of using predictive models to advance understanding of ionization efficiency and offers a valuable framework for refining SFC-MS workflows in analytical chemistry.