A deep learning-based simulator for comprehensive two-dimensional GC applications.

Area of Science:

• Analytical Chemistry
• Computational Chemistry

Background:

• Predicting eluate positions in comprehensive two-dimensional gas chromatography (GC×GC) is crucial for method optimization.
• Deep learning (DL) models offer robust and adaptable solutions for complex chromatographic data analysis.

Purpose of the Study:

• To develop an open-source deep neural network (DNN) program for optimizing GC×GC methods.
• To enable simulation of operating conditions and prediction of retention times outside the laboratory.

Main Methods:

• Development of a DNN-based program utilizing experimental predictors.
• Training and validation of DL models for predicting first- and second-dimension retention times.

Main Results:

• Achieved scaled losses (MSE) of 0.006 and 0.014 for first and second dimension retention time predictions, respectively.
• Demonstrated good prediction accuracy (R² > 0.8) for diverse chemical classes including environmental contaminants, biomolecules, and pharmaceuticals.
• Identified the need for continuous database updates for accurate prediction of less common compounds.

Conclusions:

• DNN models provide a reliable approach for retention time prediction in GC×GC.
• The developed open-source program facilitates method optimization and simulation.
• A collaborative database is proposed to enhance predictive capabilities for a wider range of compounds.