Fully automatic resolution of untargeted GC-MS data with deep learning assistance

Area of Science:

• Analytical Chemistry
• Computational Chemistry

Background:

• Co-eluting peaks in Gas Chromatography-Mass Spectrometry (GC-MS) present a significant challenge for accurate compound identification and quantification.
• Existing methods like DeepResolution require model retraining for unknown components, increasing analysis time and complexity.

Purpose of the Study:

• To develop a novel, automated pipeline (DeepResolution2) for analyzing untargeted GC-MS data that overcomes the limitations of previous methods.
• To improve the accuracy and efficiency of resolving co-eluting peaks and extracting compound-specific information.

Main Methods:

• DeepResolution2 employs deep neural networks to segment GC-MS profiles, estimate component numbers, and predict elution regions.
• These deep learning predictions guide a multivariate curve resolution procedure.
• A fixed set of seven universal models enables automated analysis of diverse GC-MS datasets.

Main Results:

• DeepResolution2 achieved superior performance compared to MS-DIAL, ADAP-GC, and AMDIS in resolving overlapped peaks, demonstrating a higher linear correlation between concentrations and peak areas (0.955 vs. 0.939, 0.948, 0.860).
• The pipeline successfully extracted peak areas and mass spectra from 136 untargeted GC-MS files of human plasma samples without manual intervention or prior information.
• DeepResolution2 provides comprehensive analysis from raw data feature extraction to discriminant model establishment.

Conclusions:

• DeepResolution2 offers a stable, universal, and automated solution for untargeted GC-MS data analysis, significantly reducing the need for manual intervention and retraining.
• The method enhances the accuracy of mass spectra, chromatograms, and peak area extraction, facilitating more reliable compound identification and quantification.
• This pipeline represents a substantial advancement in processing complex GC-MS datasets, particularly in biological and chemical research.