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

Area of Science:

• Analytical Chemistry
• Bioinformatics
• Computational Biology

Background:

• Untargeted metabolomics using liquid chromatography-high resolution mass spectrometry (LC-MS) is crucial for biological studies.
• Poor peak integration in LC-MS data preprocessing remains a significant challenge, leading to inaccurate metabolite abundance quantification.
• These inaccuracies can propagate through downstream analyses, compromising the reliability of biological insights.

Purpose of the Study:

• To develop and evaluate a computational methodology for filtering out poorly integrated peaks in untargeted metabolomics data.
• To improve the accuracy of metabolite abundance quantification by addressing a common data preprocessing bottleneck.
• To provide an automated tool for enhancing the quality of untargeted LC-MS metabolomics datasets.

Main Methods:

• Systematic comparison of 24 classifiers combining eight machine learning algorithms and three sets of peak quality metrics.
• Evaluation of classifier performance in distinguishing reliably integrated peaks from poorly integrated ones.
• Benchmarking against traditional methods like residual standard deviation (RSD) cut-offs in quality control (QC) samples.

Main Results:

• The AdaBoost algorithm combined with 11 specific peak quality metrics demonstrated superior performance in identifying and filtering unreliable peaks.
• The developed framework, when applied complementarily to data filtered by 30% RSD, further improved the removal of poorly integrated peaks.
• An R package, MetaClean, implementing the computational approach is publicly available.

Conclusions:

• The proposed machine learning-based methodology offers an effective automated solution for improving peak integration quality in untargeted LC-MS metabolomics.
• This approach enhances data reliability, leading to more accurate metabolite abundance data for downstream biological interpretation.
• The MetaClean package facilitates the widespread adoption of this advanced data filtering technique in the metabolomics community.