PeakBot: machine-learning-based chromatographic peak picking.

Area of Science:

• Metabolomics
• Analytical Chemistry
• Bioinformatics

Background:

• Chromatographic peak picking is essential for processing LC-HRMS data in untargeted metabolomics.
• Accurate peak picking is critical for comprehensive metabolic feature detection, quantification, and identification.
• Challenges include noise, overlapping peaks, and background signals that hinder accurate analysis.

Purpose of the Study:

• To develop a machine learning-based approach for robust chromatographic peak detection in LC-HRMS data.
• To improve the accuracy and reliability of peak picking in untargeted metabolomics workflows.

Main Methods:

• Developed PeakBot, a convolutional neural network-based tool for peak detection in LC-HRMS profile-mode data.
• The method extracts local signal maxima, standardizes them, and uses a custom-trained CNN for classification.
• Requires a minimum of 100 reference features for training to achieve high-quality, untargeted peak detection.

Main Results:

• PeakBot achieved high performance in discriminating chromatographic peaks from background signals, with an accuracy of 0.99.
• The tool identifies peak centers and bounding boxes, enhancing the precision of feature extraction.
• Successfully validated on independent datasets, demonstrating its effectiveness.

Conclusions:

• PeakBot offers a reliable and accurate solution for chromatographic peak picking in LC-HRMS untargeted metabolomics.
• The machine learning approach addresses limitations of traditional methods, improving data processing efficiency.
• The open-source availability facilitates its adoption in research for enhanced metabolomic analysis.