Advanced QSRR modeling of peptides behavior in RPLC

Area of Science:

• Computational Chemistry
• Chemometrics
• Chromatography

Background:

• Quantitative Structure-Retention Relationships (QSRR) model chromatographic retention using molecular descriptors.
• Large datasets necessitate variable selection for optimal QSRR model performance.
• Interpretability of descriptors versus predictive power remains a key consideration in QSRR.

Purpose of the Study:

• To compare different QSRR modeling methodologies and molecular descriptors for peptide retention prediction.
• To evaluate the predictive performance and identify important descriptors across various QSRR approaches.
• To assess the utility of interpretable versus non-interpretable descriptors in QSRR.

Main Methods:

• Employed multiple linear regression (MLR), partial least squares (PLS), uninformative variable elimination partial least squares (UVE-PLS), and genetic algorithms (GA).
• Compared predictive performance of models built using different variable selection strategies.
• Analyzed the importance of selected 1D, 2D, 3D, and empirical descriptors for peptide retention.

Main Results:

• Stepwise MLR and UVE-PLS demonstrated superior predictive performance compared to other methods.
• Selected descriptors indicated that hydrogen-bonding properties, molecular size, and complexity are crucial for RPLC retention.
• Empirical QSRR models provided the best predictions for peptide retention within the studied dataset.

Conclusions:

• Variable selection is critical for building accurate QSRR models for peptide retention.
• UVE-PLS and stepwise MLR are effective methods for QSRR model development.
• Empirical QSRR models, incorporating descriptors of molecular properties, offer the highest predictive accuracy for peptide retention.