Evaluation of normalization strategies for mass spectrometry-based multi-omics datasets.

Area of Science:

• Multi-omics data integration
• Bioinformatics
• Systems biology

Background:

• Data normalization is critical for accurate multi-omics integration, reducing systematic errors and enhancing biological signal detection.
• Existing studies often overlook time-course data, where normalization can potentially bias temporal trend analysis.
• This research addresses the need for robust normalization strategies in temporal multi-omics studies using integrated experimental designs.

Purpose of the Study:

• To establish a clear methodology for evaluating normalization impacts on multi-omics datasets.
• To identify the most reliable normalization techniques for time-course multi-omics data.
• To compare conventional normalization methods against a machine learning approach (SERRF).

Main Methods:

• Analysis of metabolomics, lipidomics, and proteomics data from human cardiomyocytes and motor neurons.
• Exposure of cells to acetylcholine-active compounds over a time course.
• Evaluation of normalization methods based on quality control (QC) feature consistency and variance analysis (treatment and time effects).

Main Results:

• Probabilistic Quotient Normalization (PQN) and LOESS QC demonstrated optimal performance for metabolomics and lipidomics.
• PQN, Median, and LOESS normalization proved effective for proteomics.
• These selected methods consistently improved QC metrics and preserved crucial biological variance (temporal or treatment-related).

Conclusions:

• PQN and LoessQC are recommended for metabolomics and lipidomics normalization in temporal multi-omics integration.
• PQN, Median, and Loess normalization are suitable for proteomics in similar contexts.
• The study provides a framework for selecting robust normalization methods in time-course multi-omics research.