BERNN: Enhancing classification of Liquid Chromatography Mass Spectrometry data with batch effect removal neural...

Area of Science:

• Proteomics and Metabolomics
• Bioinformatics and Computational Biology
• Analytical Chemistry

Background:

• Liquid Chromatography Mass Spectrometry (LC-MS) is vital for biological sample profiling.
• Batch effects from experimental variations compromise data reproducibility and interpretability.
• Existing batch correction methods often reduce biological variability.

Purpose of the Study:

• To introduce Batch Effect Removal Neural Networks (BERNN) for effective batch effect correction in large LC-MS datasets.
• To enhance sample classification performance between conditions while ensuring model generalization to unseen batches.
• To evaluate BERNN against existing methods and assess the impact on biological variation.

Main Methods:

• Development of a suite of Batch Effect Removal Neural Networks (BERNN).
• Training and testing BERNN models on diverse LC-MS datasets.
• Comparative analysis of BERNN with other batch effect correction techniques.
• Assessment of classification performance and biological variability preservation.

Main Results:

• BERNN models demonstrated superior sample classification performance across multiple datasets.
• The best classification model did not always yield the most effective batch effect removal.
• Overcorrection of batch effects led to the loss of critical biological information.

Conclusions:

• BERNN offers a promising approach for mitigating batch effects in LC-MS data.
• Balancing batch effect removal with biological variability preservation is essential for robust omics research.
• Optimizing batch correction strategies is crucial for maximizing the utility of large-scale LC-MS experiments.