Highly automatic and universal approach for pure ion chromatogram construction from liquid chromatography-mass...

Area of Science:

• Analytical Chemistry
• Computational Biology
• Biotechnology

Background:

• Liquid chromatography-mass spectrometry (LC-MS) is crucial for analyzing complex biological samples.
• Traditional feature extraction methods for LC-MS data are often manual, time-consuming, and require dataset-specific parameter optimization.
• Pure ion chromatogram (PIC) methods offer advantages over extracted ion chromatograms (EIC) by avoiding peak splitting.

Purpose of the Study:

• To develop an automated, deep learning-based method for pure ion chromatogram (PIC) extraction from LC-MS data.
• To integrate this method into an existing pipeline for comprehensive metabolomics data analysis.
• To evaluate the performance of the new method against established feature extraction tools.

Main Methods:

• A customized U-Net deep learning model was developed for direct PIC identification from centroid mode LC-MS data.
• The DeepPIC model was trained, validated, and tested using the Arabidopsis thaliana dataset.
• DeepPIC was integrated into the KPIC2 software, creating an end-to-end processing pipeline.

Main Results:

• DeepPIC demonstrated superior performance compared to XCMS, FeatureFinderMetabo, and peakonly in terms of recall rates and correlation with sample concentrations.
• Evaluation across five diverse datasets showed high precision, with 95.12% of identified PICs accurately matching manual labels.
• The KPIC2+DeepPIC pipeline provides an automatic and practical solution for feature extraction directly from raw LC-MS data.

Conclusions:

• DeepPIC offers a significant advancement in automated feature extraction for LC-MS metabolomics.
• The KPIC2+DeepPIC combination provides a robust, efficient, and universally applicable tool exceeding traditional methods.
• This automated approach facilitates large-scale, objective LC-MS data analysis.