WaveICA

WaveICA removes batch effects from large-scale untargeted metabolomics LC-MS datasets to recover biological signals for downstream analyses.

Key Features:

• Wavelet transform with ICA: Combines wavelet transform and independent component analysis (ICA) to decompose data and isolate batch-related components.
• Injection-order time trend: Leverages the time trend of samples based on injection order to identify temporal systematic biases.
• Multi-scale decomposition: Decomposes original data into multi-scale components that highlight features at different resolutions for targeted correction.
• Batch effect extraction and removal: Extracts and eliminates batch-effect information from multi-scale representations to produce corrected datasets.
• PCA clustering improvement: Improves clustering of quality control samples (QCS) and subject samples within principal component analysis (PCA) score plots.
• Pearson correlation improvement: Increased average Pearson correlation coefficients for all peaks of QCS from 0.872 to 0.972 after correction.
• Classification performance: Enhances classification accuracy and reported performance compared with three other representative methods.
• WaveICA 2.0 no-label capability: Removes batch effects without requiring explicit batch information or labels.
• Single-batch and intensity-drift correction: Demonstrates superior performance on single-batch datasets by reducing intensity drift and uncovering additional biological information.
• Implementation: Provided as an R package, WaveICA_2.0, for computational application to metabolomics data.

Scientific Applications:

• Preprocessing LC-MS untargeted metabolomics: Removes non-biological systematic biases from LC-MS untargeted metabolomics datasets prior to downstream analysis.
• Biomarker discovery: Reduces batch-related confounding to improve reliability of biomarker identification.
• Disease pathogenesis and classification studies: Enhances detection of biological signals for disease mechanism studies and supervised classification.
• Unknown-batch or single-batch studies: Applicable when batch labels are unavailable or datasets are collected as a single batch, enabling batch-effect correction in those scenarios.
• Multivariate analyses: Supports more reliable PCA and classification analyses by providing corrected input data.

Methodology:

Apply wavelet transform to decompose LC-MS data into multi-scale components, use independent component analysis (ICA) to identify batch-related components leveraging injection-order time trends, and extract/remove those components; WaveICA 2.0 implements the same approach without requiring explicit batch labels.