2025/12/24 by Roberto Laganà Vinci, Katia Arena, Francesca Rigano +4 · 1 voice
Chemistry · Computer Science · #Analytical Chemistry and Chromatography #Chromatography in Natural Products #Computational Drug Discovery Methods
paper · doi:10.1016/j.greeac.2025.100321
openalex created_date 2025/12/24 · openalex publication_date 2025/12/24 · openalex updated_date 2026/06/16
• A capillary scale LC method was developed for the separation of 53 phenolics • QSRR based on PLS regression model allowed to reliably predict retention data • Seventeen molecular descriptors were finely selected to minimize prediction errors • QSRR models showed robustness across three different injection modes • Linear retention index normalized retention data and allowed model transferability Quantitative Structure-Retention Relationship (QSRR) model was developed to predict retention data of phenolic compounds and elucidate their retention mechanisms under reversed phase conditions. Compared to a previous work (Laganà-Vinci et al. , J. Chromatogr. A, 2024, 1730, 465146), the chromatographic method was down-scaled to capillary liquid chromatography (cap-LC) scale, thus drastically reducing solvent consumption and obtaining a more eco-sustainable method with respect to conventional LC. Moreover, a linear gradient of the mobile phase was applied rather than a multi-step one, allowing the application of linear regression models (i.e. PLS (partial least squares)) and reducing overfitting risk. A dataset of 53 standard phenolic compounds was used for model building and validation, while an external set consisting in bergamot juice analytes was used to evaluate the goodness of the QSRR model. To face the challenging QSRR step of variable selection, different methods were evaluated to reduce the number of variables needed to build the model. The final selected variables were evaluated for their impact on retention mechanisms, and the robustness of PLS model was tested across three different injection methods allowed by instrument manufacturer, thus simulating model transferability between different instrumental setup. Linear retention indices (LRIs) were applied for retention data normalization to further enhance the robustness of QSRR models. The homologue series of 1-nitroalkanes was used for calculation of LRIs, taking advantage of the efficient coverage of the elution space from small and polar simple phenols to larger and medium polar structures.