ZODIAC: database-independent molecular formula annotation using Gibbs sampling reveals unknown small molecules

Marcus Ludwig; Louis-Félix Nothias; Kai Dührkop; Irina Koester; Markus Fleischauer; Martin A. Hoffmann; Daniel Petras; Fernando Vargas; Mustafa Morsy; Lihini Aluwihare; Pieter C. Dorrestein; Sebastian Böcker

The likelihood P D(c(v)) | v of a molecular formula candidate v can be computed from the posterior probability of the fragmentation tree and the isotope pattern analysis as estimated by SIRIUS 4.0 10,24 . For the Gibbs sampler, we treat these probabilities as likelihoods, although the analysis SIRIUS 4.0 also integrates certain priors 24 . To avoid proliferating running times, we usually limit further computations to the, say, 50 best-scoring molecular formulas for each compound. For each compound, we also introduce a node representing molecular formula not identied which receives likelihood from the remaining molecular formulas, and is not connected to any other nodes.