Optimal transport for automatic alignment of untargeted metabolomic data

TL;DR

GromovMatcher is introduced, a flexible and user-friendly algorithm that automatically combines LC-MS datasets using optimal transport and delivers superior alignment accuracy and robustness compared to existing approaches, and scales to thousands of features requiring minimal hyperparameter tuning.

Abstract

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

Figures (11)

• Figure 1: Performance of metabCombiner with the different parameter settings. The first setting, labelled 'Scores' correspond to the design of our main analysis, where 100 randomly selected true pairs are supplied to metabCombiner to set the scoring weights automatically, but are not otherwise used. In the second setting, labelled 'Scores + RT', metabCombiner is allowed to use the 100 true pairs not only to set the scoring weights, but also to estimate the RT drift. Finally, in the third 'Default' setting, we do not use any prior knowledge for the RT drift estimation and keep the scoring weights' default values.
• Figure 1: Sensitivity of thresholded GromovMatcher (GMT) to feature overlap fraction $\lambda$, feature imbalance fraction $\lambda_f$, and sample imbalance fraction $\lambda_s$ between two datasets being matched.
• Figure 1: Overlap between the 706 features common to the HCC and PC studies found via reference matching, and the 938 features common to HCC and PC found by direct matching
• Figure 2: Sensitivity of M2S to feature overlap fraction $\lambda$, feature imbalance fraction $\lambda_f$, and sample imbalance fraction $\lambda_s$ between two datasets being matched.
• Figure 2: Overlap between the features identified as common to the three EPIC studies using either the CS study or the HCC study as a reference.