MAVEN2: An Updated Open-Source Mass Spectrometry Exploration Platform

Abstract

MAVEN, an open-source software program for analysis of LC-MS metabolomics data, was originally released in 2010. As mass spectrometry has advanced in the intervening years, MAVEN has been periodically updated to reflect this advancement. This manuscript describes a major update to the program, MAVEN2, which supports LC-MS/MS analysis of metabolomics and lipidomics samples. We have developed algorithms to support MS/MS spectral matching and efficient search of large-scale fragmentation libraries. We explore the ability of our approach to separate authentic from spurious metabolite identifications using a set of standards spiked into water and yeast backgrounds. To support our improved lipid identification workflow, we introduce a novel in-silico lipidomics library covering major lipid classes and compare searches using our novel library to searches with existing in-silico lipidomics libraries. MAVEN2 source code and cross-platform application installers are freely available for download from GitHub under a GNU permissive license [ver 3], as are the in silico lipidomics libraries and corresponding code repository.

Keywords: GUI; fragmentation; identification; lipidomics; metabolomics; open-source; software; visualization.

Figure 1

Overview of MS/MS-based workflow in MAVEN2. MAVEN2 implements MS/MS-based slicing, construction of consensus spectra, and spectral library matching. Outline of key steps and novel algorithmic implementations are highlighted in blue (A). MS/MS scans collected in LC/MS run are used as seeds for formation of “slices”—blocks in m/z and retention time (RT) space surrounding the MS/MS scan’s precursor m/z (B). Slices from all samples are merged based on overlaps in m/z or RT space (C). These merged slices are used to generate extracted ion chromatograms (EICs) and summed to form a merged EIC. Peak groups are defined by overlapping regions of intensity in merged EIC. The peaks picked from individual sample EICs are then associated with their corresponding peak groups (D). MS/MS Spectra corresponding to a peak group are combined to form a consensus spectrum, which is then searched against spectral libraries to identify compounds (E). This approach allows for annotation of peaks in samples even when no MS/MS spectra were collected.

Figure 2

MAVEN2 User Interface. The MAVEN2 user interface presents all relevant information for compound validation in a single view. This view includes a list of loaded samples (A), overlayed EICs from multiple samples (B), sample-specific quantitation information (C), the result of a fragmentation spectral library search (D), a list of MS/MS events associated with a putative identification (E), and spectral matches between a consensus MS/MS spectrum and the library spectrum (F). Shown above is a match to a library spectrum of glutathione disulfide.

Figure 3

Metabolomics Spike-In Standards Precision Curve. True positive rate as a function of score threshold was assessed in either water or yeast backgrounds. At each threshold of MS/MS score we calculated the fraction of matches that were correctly matched (based on known retention time of spiked-in standards, see Figure S3). To allow for direct comparison between different scoring methods, the X-axis is scaled to the maximum value of each scoring method. As expected, precision was worse in the yeast background compared to the water background due to the presence of a higher number of background peaks. Performance of all methods improved with higher thresholds, with the hypergeometric score having the best performance (in terms of precision).

Figure 4

Summary of identified lipids by lipid class, dataset, and library. Counts of annotated features organized by lipid class, instrument dataset, and in-silico library. “Both” refers to features annotated as the same class by both CalicoLipids and MS-DIAL libraries. “Calico” and “MS-DIAL” refer respectively to features annotated exclusively by one library. The ten most commonly annotated classes are shown, along with all other classes represented in the “other” category. For example, the CalicoLipids library identifies many more SMs in both the internal Thermo and external Agilent datasets (SM, blue, upper and lower plots), while MS-DIAL identifies more TGs in the external Agilent dataset (TG, green lower plot).