MetaClean

MetaClean filters low-quality peak integrations from untargeted metabolomics data generated by liquid chromatography high-resolution mass spectrometry (LC-MS) to improve the accuracy of metabolite abundance estimates.

Key Features:

• Machine learning framework: Integrates machine learning with peak quality metrics to distinguish reliably integrated peaks from poorly integrated peaks.
• Classifier evaluation: Systematically evaluates 24 classifiers generated by combining eight classification algorithms with three sets of peak quality metrics.
• Optimal classifier selection: Identifies the AdaBoost algorithm combined with a specific set of 11 peak quality metrics as the most effective classifier.
• Complementary filtering: Applies classification to peaks retained after a 30% residual standard deviation (RSD) cut-off across pooled quality-control samples to identify poorly integrated peaks.
• Automated processing: Performs automated removal of unreliable peak integrations from untargeted LC-MS metabolomics datasets.

Scientific Applications:

• Metabolomics data quality: Improves the quality and reliability of untargeted LC-MS metabolomics data by filtering poorly integrated peaks.
• False positive reduction: Reduces false positives in peak detection that can distort downstream analyses.
• Biomarker discovery: Supports more robust biomarker discovery by improving metabolite quantitation.
• Metabolic pathway elucidation and systems biology: Enhances accuracy of metabolic pathway analyses and systems biology studies through more reliable metabolite abundance estimates.

Methodology:

Evaluates 24 classifiers (8 algorithms × 3 peak-metric sets) using machine learning, selects AdaBoost combined with 11 peak quality metrics, and applies the classifier to peaks retained after a 30% RSD cut-off across pooled quality-control samples to remove poorly integrated peaks.