Improved batch correction in untargeted MS-based metabolomics.

Area of Science:

• Metabolomics
• Analytical Chemistry
• Bioinformatics

Background:

• Untargeted metabolomics experiments, particularly those using mass spectrometry (MS), are prone to batch effects.
• These effects necessitate robust batch correction methods to ensure comparable peak intensities across different experimental runs.
• Non-detects (signals below detection limits) are common and must be carefully considered in correction strategies.

Purpose of the Study:

• To compare various batch correction methods for untargeted metabolomics data.
• To evaluate the impact of different strategies for handling non-detects on batch correction efficacy.
• To assess correction performance using quality criteria on large-scale LC-MS and GC-MS datasets.

Main Methods:

• Batch correction methods evaluated include regression models and normalization approaches.
• The study utilized quality control (QC) samples and randomized study samples for model fitting.
• Non-detects were handled using various imputation strategies, with a focus on their impact.

Main Results:

• Batch correction strategies incorporating batch and injection order information generally yielded superior results.
• Effective correction was achieved with normalization methods when sufficient QC samples were available.
• The method of handling non-detects significantly influenced correction quality; replacing them with zero or very small values proved suboptimal.

Conclusions:

• Quality control samples are effective for batch correction when used in sufficient numbers.
• Batch correction using study samples, when experiments are well-designed, offers comparable quality and corrects more metabolites.
• Careful selection of non-detect handling strategies is critical for achieving optimal batch correction in metabolomics.