A benchmark study of data normalisation methods for PTR-TOF-MS exhaled breath metabolomics

Area of Science:

• Volatilomics and metabolomics
• Real-time breath analysis using mass spectrometry

Background:

• Volatile organic compounds (VOCs) in exhaled breath offer diagnostic potential.
• Real-time mass spectrometry (MS), particularly proton transfer reaction (PTR) MS, is used for breath analysis.
• Data normalization is crucial to mitigate non-biological variations like batch effects and time-dependent drift in PTR-MS data.

Purpose of the Study:

• To benchmark existing metabolomic data normalization methods for real-time breath analysis.
• To assess the impact of normalization on diagnostic performance for COVID-19 using PTR-MS data.

Main Methods:

• Compared seven normalization methods (five statistical, two using internal standards) on two COVID-19 clinical trial datasets.
• Evaluated feature selection for standard metabolites and used ambient air measurements for training.
• Applied normalization to proton transfer reaction mass spectrometry (PTR-MS) data.

Main Results:

• Normalization methods successfully corrected for time-dependent drift in PTR-MS data.
• Probabilistic quotient normalization and normalization using optimal internal standards showed the best performance.
• Normalization significantly improved machine learning model performance, enhancing sensitivity, specificity, and ROC AUC for COVID-19 diagnosis.

Conclusions:

• Appropriate data normalization is essential for processing PTR-MS breath analysis data.
• Normalization significantly enhances the predictive performance of statistical and machine learning models.
• This approach improves the reliability and accuracy of breathomics for medical diagnostics.