Statistical analysis of isocratic chromatographic data using Bayesian modeling.

Area of Science:

• Analytical Chemistry
• Computational Chemistry
• Statistical Modeling

Background:

• Traditional chromatographic retention time models analyze analytes individually, limiting prediction generalization.
• Lack of shared information between analytes leads to poor out-of-sample predictions.

Purpose of the Study:

• To demonstrate the benefits of pooling data and employing a Bayesian hierarchical approach for simultaneous analysis of chromatographic data.
• To improve the generalization and accuracy of chromatographic retention time predictions.

Main Methods:

• Utilized a publicly available dataset of 1026 analytes.
• Applied a Bayesian hierarchical approach using the Stan program coupled with R for Markov chain Monte Carlo (MCMC) sampling.
• Employed the Neue model to describe retention factor (k) as a function of acetonitrile content, incorporating log P and pK_a as predictors.

Main Results:

• The analysis revealed two distinct analyte clusters based on their degree of dissociation, influencing retention behavior.
• The Bayesian model successfully incorporated prior knowledge, accounted for between-analyte variability and parameter correlations, and explained variability using predictors.
• The model demonstrated good calibration with the data, providing uncertainty quantification through posterior probability distributions.

Conclusions:

• Simultaneous analysis of chromatographic data using a Bayesian hierarchical approach enhances model performance and predictive power.
• The developed model offers insights into analyte behavior in chromatographic columns and enables predictions for diverse analytes based on log P and pK_a values.
• This approach is particularly valuable when dealing with limited data for specific model parameters, facilitating information sharing across analytes.