XY-Meta: A High-Efficiency Search Engine for Large-Scale Metabolome Annotation with Accurate FDR Estimation

Abstract

FDR control has been a huge challenge for large-scale metabolome annotation. Although recent research indicated that the target− decoy strategy could be implemented to estimate FDR, it is hard to perform FDR control due to the difficulty of getting a reliable decoy database because of the complex fragmentation mechanism of metabolites and ubiquitous isomers. To tackle this problem, we developed a decoy generation method, which generates forged spectra from the reference target database by preserving the original reference signals to simulate the presence of isomers of metabolites. Benchmarks on GNPS data sets in Passatutto showed that the decoy database generated by our method is closer to the actual FDR than other methods, especially in the low FDR range (0−0.05). Large-scale metabolite annotation on 35 data sets showed that strict FDR reduced the number of annotated metabolites but increased the spectral efficiency, indicating the necessity of quality control. We recommended that the FDR threshold should be set to 0.01 in large-scale metabolite annotation. We implemented decoy generation, database search, and FDR control into a search engine called XY-Meta. It facilitates large-scale metabolome annotation applications.

Modifying Chromatography Conditions for Improved Unknown Feature Identification in Untargeted Metabolomics

“…Compound identification accuracy could also be improved by other strategies. Methods to estimate false discovery rates (FDRs) in metabolomics have been proposed − and could dramatically improve the ability to discern correct from incorrect database hits, but these approaches are not yet reliable enough to be widely used. Alternatively, biological “constraints” could be used to assess the probability of detecting a specific metabolite in a particular sample type.…”

Molecular formula discovery via bottom-up MS/MS interrogation

“…Efforts have been made to control FDR on MS/MS spectral matching-based metabolite identification 27, 35, 46 , yet the FDR estimation for unknown annotation 10 remains elusive. In our approach, Platt scaling monotonically maps MLR prediction scores onto a probability scale.…”

Advances in decomposing complex metabolite mixtures using substructure- and network-based computational metabolomics approaches

“…The most pressing is the relatively small size of the reference databases: only a very small subset of known chemicals are covered. In addition, although work has been done in this area, [55][56][57][58] there is still no widely used method for computing false discovery rates for these data, which can make interpretation of cosine scores highly subjective without extensive manual inspection of spectral matches in terms of overlapping peaks and their intensity patterns.…”