Deep Learning for the Precise Peak Detection in High-Resolution LC-MS Data

Area of Science:

• Computational Biology
• Analytical Chemistry

Background:

• Metabolomics data analysis relies on accurate peak detection and integration in liquid chromatography-mass spectrometry (LC-MS).
• Existing algorithms often exhibit poor precision, leading to false positive signals and reduced data reliability.

Purpose of the Study:

• To develop a novel machine learning algorithm for precise peak detection and integration in LC-MS data.
• To improve the accuracy of initial data processing steps in metabolomics.

Main Methods:

• Application of convolutional neural networks, a type of machine learning.
• Development of a flexible algorithm named 'peakonly' for selective detection and exclusion of noisy peaks.
• Testing on high-resolution LC-MS data for metabolomics.

Main Results:

• The 'peakonly' algorithm demonstrates high flexibility in handling low-intensity noisy peaks.
• Achieves excellent quality in true positive peak detection, approaching maximal precision.
• Significantly reduces false positive signals compared to widely used algorithms.

Conclusions:

• The 'peakonly' algorithm offers a significant advancement in metabolomics data analysis precision.
• The approach is suitable for high-resolution LC-MS data and potentially adaptable for GC-MS.
• The tool is freely available, promoting wider adoption and further research.