Optimal transport for automatic alignment of untargeted metabolomic data

Area of Science:

• Metabolomics
• Analytical Chemistry
• Bioinformatics

Background:

• Untargeted metabolomic profiling using liquid chromatography-mass spectrometry (LC-MS) is crucial for drug development, disease diagnosis, and risk prediction.
• Merging multiple LC-MS datasets is essential for robust biomarker discovery but is hindered by low throughput and data variations.
• Existing data pooling methods face limitations due to sensitivity to data variations and reliance on hyperparameters.

Purpose of the Study:

• To introduce GromovMatcher, a novel algorithm for automated merging of LC-MS datasets.
• To enhance the accuracy and robustness of data alignment in untargeted metabolomics.
• To facilitate biomarker discovery by enabling effective integration of diverse LC-MS data.

Main Methods:

• Developed GromovMatcher, an algorithm utilizing optimal transport for automatic LC-MS dataset merging.
• Leveraged feature intensity correlation structures for superior alignment accuracy and robustness.
• Created a dataset split procedure for generating validation pairs to assess alignment algorithms.

Main Results:

• GromovMatcher demonstrated superior alignment accuracy and robustness compared to existing methods.
• The algorithm effectively scales to thousands of features with minimal hyperparameter tuning.
• Applied to patient studies, GromovMatcher identified metabolic features associated with alcohol intake in liver and pancreatic cancer.

Conclusions:

• GromovMatcher offers a flexible, user-friendly solution for combining LC-MS datasets, overcoming limitations of current approaches.
• The algorithm facilitates the discovery of biomarkers linked to lifestyle risk factors, such as alcohol consumption in cancer.
• This method advances untargeted metabolomics for biomarker discovery in complex biological and clinical studies.