ProbMetab: an R package for Bayesian probabilistic annotation of LC-MS based metabolomics

• Published in: arXiv.org
• Main Authors: Silva, Ricardo R, Jourdan, Fabien, Salvanha, Diego M, Letisse, Fabien, Jamin, Emilien L, Guidetti-Gonzalez, Simone, Labate, Carlos A, Vêncio, Ricardo Z N
• Format: Paper Journal Article
• Language: English
• Published: Ithaca Cornell University Library, arXiv.org 27.10.2013
• Subjects: Annotations; Bayesian analysis; Biological effects; Correlation analysis; Dirichlet problem; Metabolites; Probabilistic inference; Quantitative Biology - Genomics; Quantitative Biology - Quantitative Methods; Substrates
• ISSN: 2331-8422
• DOI: 10.48550/arxiv.1310.7226

We present ProbMetab, an R package which promotes substantial improvement in automatic probabilistic LC-MS based metabolome annotation. The inference engine core is based on a Bayesian model implemented to: (i) allow diverse source of experimental data and metadata to be systematically incorporated into the model with alternative ways to calculate the likelihood function and; (ii) allow sensitive selection of biologically meaningful biochemical reactions databases as Dirichlet-categorical prior distribution. Additionally, to ensure result interpretation by system biologists, we display the annotation in a network where observed mass peaks are connected if their candidate metabolites are substrate/product of known biochemical reactions. This graph can be overlaid with other graph-based analysis, such as partial correlation networks, in a visualization scheme exported to Cytoscape, with web and stand alone versions. ProbMetab was implemented in a modular fashion to fit together with established upstream (xcms, CAMERA, AStream, mzMatch.R, etc) and downstream R package tools (GeneNet, RCytoscape, DiffCorr, etc). ProbMetab, along with extensive documentation and case studies, is freely available under GNU license at: http://labpib.fmrp.usp.br/methods/probmetab/.