Automated component analysis in DOSY NMR using information criteria

Area of Science:

• Analytical Chemistry
• Physical Chemistry
• Spectroscopy

Background:

• Diffusion-Ordered Spectroscopy (DOSY) is crucial for analyzing complex mixtures.
• Accurate estimation of mixture components in DOSY NMR is challenging.
• Model selection techniques are vital for reliable data interpretation.

Purpose of the Study:

• To introduce a model selection technique using Bayesian information criteria for DOSY NMR.
• To evaluate the performance of Weighted Least Squares (WLS) and Maximum a Posteriori (MAP) estimators.
• To determine the accuracy and limitations of these estimators in mixture analysis.

Main Methods:

• Bayesian information criteria for model selection.
• Weighted Least Squares (WLS) with L2-regularization.
• Maximum a Posteriori (MAP) with prior information database.
• Data resampling techniques for error estimation.

Main Results:

• WLS effectively detects components with >2-fold diffusion coefficient differences.
• MAP improves precision, detecting components with 1.5-fold differences.
• Both estimators yield weight ratio estimates with ~1% standard deviation.
• MAP may overestimate components with larger self-diffusion coefficients.

Conclusions:

• The developed Bayesian criteria enhance DOSY NMR component estimation.
• WLS and MAP estimators offer distinct advantages in accuracy and precision.
• Further refinement is possible based on examined data distribution characteristics.