Significance estimation for large scale metabolomics annotations by spectral matching

Abstract

The annotation of small molecules in untargeted mass spectrometry relies on the matching of fragment spectra to reference library spectra. While various spectrum-spectrum match scores exist, the field lacks statistical methods for estimating the false discovery rates (FDR) of these annotations. We present empirical Bayes and target-decoy based methods to estimate the false discovery rate (FDR) for 70 public metabolomics data sets. We show that the spectral matching settings need to be adjusted for each project. By adjusting the scoring parameters and thresholds, the number of annotations rose, on average, by +139% (ranging from −92 up to +5705%) when compared with a default parameter set available at GNPS. The FDR estimation methods presented will enable a user to assess the scoring criteria for large scale analysis of mass spectrometry based metabolomics data that has been essential in the advancement of proteomics, transcriptomics, and genomics science.

High-Resolution Liquid Chromatography Tandem Mass Spectrometry Enables Large Scale Molecular Characterization of Dissolved Organic Matter

“…This approach paired with high acquisitions speed (>1 Hz) of state of the art instruments results in thousands of spectra per LC-MS/MS run. For a reliable data analysis and reproducible interpretation of the results, bioinformatic workflows including comprehensive databases and statistical significance estimation are crucial (da Silva et al, 2015;Böcker, 2017;Scheubert et al, 2017;Weber et al, 2017) and have been very recently employed for marine metabolomic studies Hartmann et al, 2017;Kujawinski et al, 2017;Longnecker and Kujawinski, 2017). With these new bioinformatic tools and instrumental improvements in sensitivity, acquisition speed and resolution we anticipate that the techniques used for DOM characterization will further shift toward non-targeted analyses using high-resolution LC-MS/MS that provide inventories of molecular structures in complex environmental datasets.…”

Application of Global Metabolomics to the Identification of Complex Counterfeit Medicinal Products

“…It is also worth noting that the estimation of false discovery rates, has been a historical goal 17 and recent focus for small molecule identification. 18,19 However, these tools are not presently in mainstream use and do not exist in the software used for this study. The mzCloud library does possess a sophisticated scoring mechanism for quality of MS/MS spectra as shown in Figure 1B for the amino acid asparagine, but the other libraries rely on MS1 mass accuracy alone.…”

MESSAR: Automated recommendation of metabolite substructures from tandem mass spectra

“…Interestingly, 290 the MESSAR-predicted substructures showed a striking similarity to expert knowledge, 291 ranging from simple (e.g ethyl phenol of Motif 21) to complex (e.g. indole substructure 292 of Motif 25,26,194) substructures. According to experts' knowledge, ground-truth 293 annotations of 26 motifs (out of 28) were identical or very similar to the substructure 294 predicted by at least one matched MESSAR rule.…”

“…Training spectral libraries 75 MESSAR generates rules from target and decoy GNPS spectral libraries built by 76 Scheubert et al (Fig 1A). According to the data descriptions [26], the target library 77 consists of 4138 positive ion high-quality labeled spectra acquired on Q-TOF instru-78 ments. For each spectrum, Scheubert et al have computed a fragmentation tree that 79 annotates a subset of fragments with molecular formulas and removed non-annotated 80 peaks that usually represent isotopic peaks, chemical noise, .…”

“…The algorithm discovers rules among labeled spectra based 200 on user-defined parameters (S1 Text). An overview of the rule generation procedure 201 from the spectral library is depicted in Fig 1. This includes the following steps described 202 in the Materials and Methods section: (A) retrieving training data: target and decoy 203 GNPS spectral libraries used by passatutto software [26], (B) extracting spectral features 204 from the target library, (C) generating molecular substructures of training compounds, 205 (D) extracting spectral features from the decoy library, (E) ARM on target features and 206 substructures, (F) generating random rules from the decoy database , (G) statistical 207 evaluation of target and decoy rules, filtering target rules by FDR estimation with the 208 target-decoy method. All target rules above the score threshold chosen were saved in 209 the MESSAR rule database.…”