Machine Learning-Based Retention Time Prediction Tool for Routine LC-MS Data Analysis.

Area of Science:

• Analytical Chemistry
• Computational Chemistry
• Cheminformatics

Background:

• Liquid chromatography-mass spectrometry (LC-MS) is crucial for chemical synthesis analysis.
• Accurate retention time (RT) prediction is essential for improving LC-MS data processing.
• Large internal datasets from chemical synthesis provide opportunities for model development.

Purpose of the Study:

• To develop and evaluate a novel RT prediction model for LC-MS data analysis.
• To leverage internal experimental data and advanced neural network architectures.
• To enhance the analytical capabilities of in-house LC-MS workflows.

Main Methods:

• Development of a graph neural network (NN) model using GATv2Conv + DL architecture.
• Training the NN model on a large internal dataset of chemical synthesis experiments.
• Evaluation of the model using the METLIN SMRT dataset and a 120 s LC-MS method.

Main Results:

• The developed RT prediction model achieved a mean absolute error (MAE) of 2.48 s.
• Over 95% of prediction errors fell within the interval of RT ± 7.12 s to RT ± 9.58 s.
• The model demonstrated successful integration into an in-house LC-MS analysis toolkit.

Conclusions:

• The GATv2Conv + DL graph NN model provides accurate RT predictions for LC-MS analysis.
• The model significantly enhances the predictive and analytical capabilities of chemical synthesis workflows.
• A subset of 20,000 data points is publicly released to support community research and benchmarking.