Optimal transport for automatic alignment of untargeted metabolomic data

• DOI: 10.7554/elife.91597
• Publication Date: 2023-12-11T15:53:27Z
• Authors: Marie Breeur; George Stepaniants; Pekka Keski-Rahkonen; Philippe Rigollet; Vivian Viallon
• Keywords: Pooling; Hyperparameter; Biomarker Discovery; Profiling (computer programming)

Untargeted metabolomic profiling through liquid chromatography-mass spectrometry (LC-MS) measures a vast array of metabolites within biospecimens, advancing drug development, disease diagnosis, and risk prediction. However, the low throughput LC-MS poses major challenge for biomarker discovery, annotation, experimental comparison, necessitating merging multiple datasets. Current data pooling methods encounter practical limitations due to their vulnerability variations hyperparameter dependence. Here, we introduce GromovMatcher, flexible user-friendly algorithm that automatically combines datasets using optimal transport. By capitalizing on feature intensity correlation structures, GromovMatcher delivers superior alignment accuracy robustness compared existing approaches. This scales thousands features requiring minimal tuning. Manually curated validating algorithms are limited in field untargeted metabolomics, hence develop dataset split procedure generate pairs validation test alignments produced by other methods. Applying our method patient studies liver pancreatic cancer, discover shared metabolic related alcohol intake, demonstrating how facilitates search biomarkers associated with lifestyle factors linked several cancer types.