Kernel-Based, Partial Least Squares Quantitative Structure-Retention Relationship Model for UPLC Retention Time...

Area of Science:

• Analytical Chemistry
• Computational Chemistry
• Cheminformatics

Background:

• Accurate prediction of retention times is crucial for chromatographic analysis and metabolite identification (MetID).
• Traditional quantitative structure-retention relationship (QSRR) models often face limitations in predictive accuracy and applicability across diverse chemical structures.
• Developing robust QSRR models that integrate various molecular descriptors is essential for advancing chromatographic data analysis.

Purpose of the Study:

• To introduce a novel QSRR model based on Kernel-based partial least-squares (KPLS) for predicting UPLC retention times in reversed-phase mode.
• To evaluate the performance of the KPLS model against traditional regression methods like Multiple Linear Regression (MLR) and Partial Least Squares (PLS).
• To demonstrate the model's utility in real-world MetID tasks and its adaptability to different chromatographic conditions.

Main Methods:

• Development of a KPLS model using a dataset of 1383 compounds with diverse chemical structures.
• Inclusion of both classical (physicochemical, topological) and non-classical (fingerprints) molecular descriptors.
• Random data splitting into training and test sets for rigorous model validation and comparison with MLR and PLS models.

Main Results:

• The KPLS model achieved high predictive accuracy, with the best predicted/experimental R² value exceeding 0.86 and Q² close to 0.84.
• KPLS demonstrated superior performance over MLR and PLS in terms of correlation, prediction accuracy, and support for MetID peak assignment.
• The model successfully predicted the elution order of Phase I metabolites, including isomeric compounds, in two real-life MetID applications.

Conclusions:

• The proposed KPLS-based QSRR model offers a significant advancement in predicting UPLC retention times.
• The model's ability to handle diverse chemical structures and its adaptability to varying gradient profiles provide broad flexibility for analytical applications.
• This KPLS model enhances the reliability of MetID by accurately predicting metabolite elution orders, supporting complex analytical workflows.