AI-Enhanced Understanding of Retention Interactions in Supercritical Fluid Chromatography: Neural Network Insights into Retention on Selected Non-Polar Stationary Phases

The retention behavior in supercritical fluid chromatography (SFC) remains a complex and poorly understood phenomenon despite the development of various models to explain retention mechanisms. This study aims to deepen the understanding of retention by investigating three distinct stationary phases: high-strength silica octadecyl (HSS C18 SB), charged surface hybrid pentafluorophenyl (CSH PFP), and porous graphitic carbon (PGC) as a nonsilica-based phase. Three mobile phase compositions, i.e., CO₂/methanol, CO₂/methanol +10 mmol/L NH₃, and CO₂/methanol +2% H₂O, were investigated using an extensive set of analytes characterized by over 200 molecular descriptors. Artificial neural networks were employed to analyze the influence of these descriptors on retention behavior, revealing the most significant molecular features that increase or decrease retention on each column with the three different mobile phases. This complex evaluation of the large set of experimental data enabled to link specific analyte properties to retention interactions in SFC, including the interaction of analytes with partial positive charge with silanol groups on the HSS C18 SB column when using methanol + H₂O as the organic modifier. The flexibility of the alkyl chain in the HSS C18 SB column is also affected by the composition of the organic modifier, which alters retention mechanisms, especially when NH₃ is used as an additive. This highlights the critical role of the mobile phase composition in modulating the behavior of nonpolar stationary phases. Completely different interaction mechanisms were observed for the PGC column when comparing methanol with and without additives, suggesting possible modifications to the planar structure and surface polarizability of the PGC phase. Statistical evaluation of data collected over a year of column usage demonstrated distinct long-term retention stability trends. The HSS C18 SB column exhibited the greatest stability with methanol + H₂O, whereas significant retention decreases were observed with methanol + NH₃ modifier, particularly for CSH PFP and, unexpectedly, also for PGC. These findings provide crucial insights into the long-term retention behavior and aging of SFC columns, with practical implications for optimizing SFC conditions and improving column lifetime.