Pathway-Activity Likelihood Analysis and Metabolite Annotation for Untargeted Metabolomics Using Probabilistic...

Area of Science:

• Biochemistry
• Computational Biology
• Systems Biology

Background:

• Untargeted metabolomics provides comprehensive small molecule profiling but faces challenges in interpreting complex data and determining biological pathway activity.
• Existing computational tools struggle with accurate metabolite identification and pathway analysis, limiting biological insights.

Purpose of the Study:

• To introduce Probabilistic modeling for Untargeted Metabolomics Analysis (PUMA), an inference-based computational approach for enhanced metabolomics data interpretation.
• To improve the prediction of active biochemical pathways and the probabilistic annotation of metabolite measurements.

Main Methods:

• PUMA utilizes a generative model to integrate metabolomics measurements with biological networks.
• Stochastic sampling is employed to compute posterior probability distributions for pathway activity and metabolite annotation.
• Pathway activity is defined by the probability that the pathway generated the observed measurements, exceeding a user-defined threshold.

Main Results:

• PUMA demonstrates an average 8% improvement over traditional pathway enrichment analysis on synthetic datasets.
• Case studies show PUMA identifies biologically meaningful active pathways and provides annotations consistent with spectral signature-based methods.
• PUMA successfully assigned chemical identities to 23 previously isomeric metabolites and offered numerous additional putative annotations.

Conclusions:

• PUMA offers a robust probabilistic framework for interpreting untargeted metabolomics data, enhancing both pathway activity prediction and metabolite annotation.
• The approach provides higher precision (0.833) and recall (0.676) for metabolite identification compared to existing methods.
• PUMA advances the biological interpretation of metabolomics datasets, facilitating deeper understanding of cellular processes.