MetaClean: a machine learning-based classifier for reduced false positive peak detection in untargeted LC-MS metabolomics data | JoVE Visualize

MetaClean: a machine learning-based classifier for reduced false positive peak detection in untargeted LC-MS metabolomics data

Kelsey Chetnik¹, Lauren Petrick^2,3, Gaurav Pandey^4,5

¹Department of Genetics and Genomic Sciences and Icahn Institute for Data Science and Genomic Technology, Icahn School of Medicine at Mount Sinai, New York, NY, USA.

Metabolomics : Official Journal of the Metabolomic Society|October 21, 2020

View abstract on PubMed

Summary

Poor peak integration in untargeted metabolomics data is a common issue. We developed a machine learning approach using peak quality metrics to accurately filter out unreliable metabolite peaks from LC-MS data.

Area of Science:

Metabolomics
Computational Biology
Analytical Chemistry

Background:

Untargeted metabolomics using liquid chromatography-high resolution mass spectrometry (LC-MS) is powerful but susceptible to poor peak integration.
Incorrect metabolite abundances from poor integration can negatively impact downstream analyses.
Existing pre-processing software often fails to adequately address this peak integration problem.

Purpose of the Study:

To develop and evaluate a computational methodology for filtering low-quality peaks in untargeted LC-MS metabolomics data.
To combine machine learning algorithms with peak quality metrics for improved peak integration assessment.
To provide an automated tool for enhancing the reliability of metabolomics data.

Main Methods:

Systematically compared 24 classifiers from 8 algorithms and 3 peak quality metric sets.
Evaluated classifiers against residual standard deviation (RSD) cut-offs in quality control (QC) samples.
Developed an R package, MetaClean, implementing the best performing computational approach.

Main Results:

The AdaBoost algorithm combined with 11 peak quality metrics demonstrated superior performance in identifying reliably integrated peaks.
The developed framework effectively filtered out poorly integrated peaks missed by standard RSD thresholding.
The MetaClean R package offers an accessible implementation of this advanced filtering technique.

Conclusions:

This study presents a significant advancement in automated tools for filtering unreliable peak integrations in untargeted LC-MS metabolomics.
The proposed machine learning approach enhances data quality and reliability for metabolomics studies.
MetaClean provides a valuable resource for researchers to improve the accuracy of their metabolomics data analysis.