Gradient Retention Time Modeling in Ion Chromatography through Ensemble Machine Learning-Powered Quantitative...

Area of Science:

• Analytical Chemistry
• Chromatography

Background:

• Quantitative structure-retention relationships (QSRRs) are widely used in ion chromatography to predict retention times from molecular structures.
• Existing methods often couple QSRRs with solvent strength models, which can propagate and amplify errors due to inconsistencies.

Purpose of the Study:

• To develop more accurate QSRR models by directly incorporating isocratic conditions, thereby reducing error propagation.
• To build global models that account for both global and local sources of variability in chromatographic retention.

Main Methods:

• Evaluated four machine learning approaches: random forest regression, gradient boosting regression (GBR), extreme gradient boosting (xgBoost), and adaptive boosting (AdaBoost).
• Compared these models against a baseline partial least-squares model.
• Incorporated developed QSRR models into an isocratic-to-gradient model for predicting gradient retention.

Main Results:

• GBR and xgBoost demonstrated superior predictive ability for isocratic retention, with root-mean-square errors (RMSEs) of 0.025.
• These GBR and xgBoost QSRR models also outperformed others in predicting gradient retention, achieving RMSEs of 0.358 and 0.385 min, respectively.

Conclusions:

• Directly incorporating eluent composition into QSRR models significantly reduces error propagation and improves predictive accuracy.
• The developed machine learning models, particularly GBR and xgBoost, offer a robust approach for predicting chromatographic retention under various conditions.
• This methodology holds potential for extension to other chromatographic techniques.