Single sample pathway analysis in metabolomics: performance evaluation and application.

Area of Science:

• Metabolomics
• Bioinformatics
• Systems Biology

Background:

• Single sample pathway analysis (ssPA) converts omics data to pathway level for patient-specific signature discovery.
• ssPA enables multi-group comparisons and downstream analyses like pathway-based machine learning.
• Limited literature exists on ssPA's suitability for metabolomics compared to transcriptomics.

Purpose of the Study:

• To benchmark established and novel ssPA methods for metabolomics data.
• To evaluate the performance of ssGSEA, GSVA, SVD (PLAGE), z-score, ssClustPA, and kPCA.
• To demonstrate ssPA's application in pathway-based interpretation of metabolomics data, including subtype identification.

Main Methods:

• Benchmarking of six ssPA methods (ssGSEA, GSVA, SVD (PLAGE), z-score, ssClustPA, kPCA) using semi-synthetic metabolomics data.
• Application of ssPA to inflammatory bowel disease mass spectrometry data.
• Utilizing clustering to identify subtype-specific pathway signatures and visualize correlation networks.

Main Results:

• GSEA-based and z-score methods showed higher recall.
• Clustering/dimensionality reduction methods offered greater precision at moderate-to-high effect sizes.
• ssPA provided richer interpretation of inflammatory bowel disease data, enabling visualization of pathway-based patient subtype networks.

Conclusions:

• ssPA methods significantly enhance the value of metabolomic studies.
• This work serves as a reference for applying ssPA to metabolomics.
• A Python package (sspa) is available for implementing the benchmarked methods.