Advanced liquid chromatography techniques and quantitative structure-retention relationship approach for phenolic compounds analysis: a journey toward greener and reliable analytical methods

In the -omics field, phenolic compounds remain analytically demanding for liquid chromatography (LC) due to their abundance, structural heterogeneity, and isomerism. While advances in HPLC, miniaturised formats (nano-/capillary-/micro-LC) and comprehensive two-dimensional LC (LC×LC), coupled with PDA and MS (or MS/MS and HRMS) detection, have broadened capabilities, routine workflows still have to deal with identification uncertainty and solvent-intensive method optimisation. This thesis addresses these limitations by developing an interpretable and transferable modelling workflow grounded in Quantitative Structure–Retention Relationships (QSRR) and integrated with Linear Solvent Strength (LSS) theory. The workflow was optimised and applied across complementary LC platforms, such as conventional HPLC, capillary-LC, and LC×LC, to predict retention data, rationalise retention mechanisms, and support greener method design. Gradient retention factors and Linear Retention Indices were used as normalising variables to enable cross-system comparison and transferability, thereby helping to fill the gap between liquid and gas chromatography through retention traceability and laying the groundwork for reliable and transferable LC databases. Integration of LSS within the QSRR framework, consolidated during the visiting period at the Faculty of Chemical Technology, University of Pardubice, enabled further predictions under new theoretical gradients and conditions while maintaining mechanistic coherence. Models were intentionally constrained to be interpretable and auditable, with the final employed descriptor sets reflecting chemically meaningful information about retention and providing system￾specific insight without relying on so-called black-box optimisation, typical of neural network models. The resulting QSRR–LSS workflow showed high potential for reducing experimental burden and solvent consumption by guiding method development and improving confidence in identification by aligning predicted retention with spectral evidence, consequently facilitating knowledge transfer across instruments, columns, and injection modes. Although developed for the complex and heterogeneous class of phenolic compounds, the approach is general and should be even easier to apply to simpler compound families. It provides the basis for a scalable workflow for identification and optimisation based on retention data prediction in other analyte classes and establishes a practical route towards transferable LC databases, which could be further improved using larger training datasets and more powerful but information-blind models.