What can go wrong at the data normalization step for identification of biomarkers?

Area of Science:

• Analytical Chemistry
• Chemometrics
• Biomarker Discovery

Background:

• Instrumental signals from techniques such as High-Performance Liquid Chromatography-Diode Array Detection (HPLC-DAD), Liquid Chromatography-Mass Spectrometry (LC-MS), and Ultra-Performance Liquid Chromatography-Mass Spectrometry (UPLC-MS) often exhibit a 'size effect'.
• This size effect, stemming from variations in sample volume or concentration, prevents absolute quantification and complicates data comparison based on sample fingerprints.
• Focusing on the 'shape effect', which captures information in variable ratios, is crucial for meaningful data analysis.

Purpose of the Study:

• To evaluate and compare the effectiveness of various normalization methods in removing the size effect from instrumental data.
• To assess the performance of Compositional Data Analysis (CODA) methods, utilizing log-ratio transformations, against traditional normalization techniques.
• To determine which methods best facilitate the accurate identification of biomarkers.

Main Methods:

• Comparison of popular normalization techniques with CODA-based approaches.
• Application of log-ratio transformations inherent to CODA.
• Evaluation of method performance based on biomarker identification accuracy.

Main Results:

• The study systematically compares the efficacy of established normalization methods against CODA techniques.
• Performance evaluation focuses on the ability of each method to mitigate the size effect and preserve shape information.
• Results indicate the potential of CODA methods for improved biomarker discovery.

Conclusions:

• Normalization is essential for removing the size effect in chromatographic and mass spectrometry data.
• CODA methods, leveraging log-ratio transformations, offer a robust alternative for analyzing compositional data.
• The findings support the use of CODA for enhanced biomarker identification in complex samples.