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

Gibbs sampling is a Markov chain Monte Carlo algorithm for obtaining a sequence of observations approximated from a multivariate probability distribution 57 . Sampling assignments according to (2) . CC-BY-NC-ND 4.0 International license author/funder. It is made available under a The copyright holder for this preprint (which was not peer-reviewed) is the . https://doi.org/10.1101/842740 doi: bioRxiv preprint can be seen as an archetype application of a Gibbs sampler: We start with some assignment, such as the highest likelihood node (molecular formula) for each compound (color). Each epoch of the Gibbs sampler consists of |C| steps, where we iterate over all colors c ∈ C in random order: We update the active node with color c by drawing a node with color c according to its posterior probability, conditional the current assignment of all nodes with color dierent from c. At the end of the epoch we output the current assignment, and repeat until we have reached a sucient number of samples.