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

Finding and scoring ZODIAC anchors. ZODIAC can use (potentially incorrect) spectral library hits as anchors to improve annotations. To nd a reasonable number of anchors, we perform spectral library search in analogue mode. Resulting molecular formula annotations are not considered ground truth identications but are sucient as anchors. Only those hits were considered that have mass dierences between query and reference corresponding to a frequent biotransformation. We use the following molecular formula mass dierences as valid biotransformations We use identical parameters for all ve datasets: When scoring anchors according to (5), we use the maximum of the cosine score between the spectrum and the cosine score of the mirrored spectrum as the similarity measure, and min l = 0.5 as the score threshold parameter and λ = 1, 000 as the weighting parameter. For anchors found by spectral library match in analogue mode (that is, non-identical m/z ), spectral similarity is reduced by 0.1 to account for increased uncertainty. Searching for anchors as described above resulted in 96 anchors for dendroides, 254 anchors for NIST1950, 749 for tomato, 372 for diatoms and 176 for mice stool. All spectral hits described in the previous section are anchors, too; recall that for dendroides, the ground truth was established manually and those annotations do not serve as anchors.