Successes and pitfalls in automated dereplication strategy using liquid chromatography coupled to mass spectrometry data: A CASMI 2016 experience

Abstract

Automated annotation of data, originating from liquid chromatography coupled to high-resolution mass spectrometry profiles (LC–HRMS), remains a highly challenging task. Therefore, the Critical Assessment of Small Molecule Identification (CASMI) Contest (http://casmi-contest.org/) represents a unique opportunity to blindly evaluate annotation workflows. The 2016 CASMI contest consisted of 16 LC–HRMS/MS profiles with 18 detected peaks to annotate. Those peaks corresponded to compounds from natural origin. An R script based on the XCMS, IPO, RMassBank, CAMERA and MeHaloCoA packages was devised. Two other external tools: SIRIUS3 and CFM-ID were also integrated for molecular formulae and in silico fragmentation calculation, respectively. This script was used to perform peak picking, spectral interpretation, molecular formula determination and database search for structural determination. Finally, the structures were further discriminated based on in silico fragmentation. After the release of the CASMI contest solutions, successes and failures of the proposed script were investigated. In most cases, no differences were observed in the rank of the correct structure when using raw LC–HRMS data or manually obtained MS and MS/MS spectra. However, the study of the few cases where differences were detected tends to show that automatic detection of MS2 data within the raw LC–MS data yielded more accurate identification. The failures in proposing the correct structure within the submission list were related to the absence of the right structure in the interrogated databases. However, very close structure were proposed in first rank indicating that such approaches are able to rapidly determine the carbon skeleton of the structure; the medium rank of the correct structure in the proposed list for each peak of interest being 2nd.