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

To evaluate the performance of SIRIUS and ZODIAC, we had to annotate a subset of compounds with correct molecular formulas, to serve as our ground truth. For this, we combined manual annotation and spectral library search, as follows: For the dendroides dataset, spectral library hits were obtained for the isolated molecules that had their reference MS/MS spectra added to the GNPS library. We used molecular networking and spectral library search in analog mode 1 , along with a set of known typical biotransformation, to annotate related diterpene esters. They dier mainly by their acylation degree, and the nature of acyl residues on the diterpene backbone. This resulted in 201 compounds being annotated with molecular formulas by manual analysis of the data, see Supplementary Table 5. For the remaining datasets, we performed spectral library searches against multiple libraries, but did not add manual annotations. We searched compounds in a spectral library combining GNPS 1 , MassBank 4 , NIST17 database (National Institute of Standards and Technology, v17) and