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

Furthermore, we assume that some compounds were identied by searching in a library of tandem MS spectra, plus potentially by comparison of retention times. We refer to these compounds and the corresponding molecular formulas as anchors. Such library search results can also be wrong, so we do not exclude other molecular formula explanations, but rather give a bonus to the likelihood of the identied molecular formula. The quality of a spectral library hit can, to a certain extend, be evaluated using its score, usually the dot product (cosine score) between query and reference. Hence, the bonus may be dependent on the corresponding library search score. Given the library search score s l ∈ [0, 1] and a minimum score to consider a library hit min l , we multiply the candidate's likelihood by ψ(s l , min l ) = exp λ max(s l , min l ) 1 − max(s l , min l ) .