Abstract
LC-accurate mass MS/MS is widely used in metabolomics research, screening biological samples and comprehensively quantifying and identifying small molecules in order to gain insight into the physiological and chemical conditions of the cell. While targeted and multi-targeted methods methods give very precise data, such methods are less suitable for discovery driven research. On the other hand, data processing for untargeted methods involves complex procedures yielding hundreds to thousands of peak findings that are hard to validate. In this study, data processing methods for the two most cited open access software programs used for untargeted metabolomics data processing, XCMS Online and MZmine, were compared with MS-DIAL, an open access software program recently introduced by the RISKEN Institute and Fiehn laboratory in 2015. Results for these three open access software programs were evaluated based on their true positive rate of recovery of 14 class specific internal standards, 510 lipids in an accurate mass retention time library, and of low abundant peaks. Methods were developed and optimized for MZmine 2.10 using technical replicates of plasma lipidomic data, whereas the software developer settings were used for XCMS online and MS-DIAL. An accurate mass retention time library of lipids was constructed by manually investigating all detected ions to determine a true peak list table of 510 features. Peaks discarded from the true peak list table were peaks detected in the method blank at a peak intensity > 20 % of the samples and were not reproducible across multiple samples with a CV > 20%. Seven replicate human plasma samples were processed in Agilent Mass Hunter Quantitative Analysis B.07.00 to manually investigate and integrate the 510 features listed in the true peak list table. The peak intensities for the internal standards from Agilent Mass Hunter Quantitative Analysis were compared to XCMS Online, MZmine, and MS-DIAL and had a correlation coefficient of 0.954, 0.983, and 0.999 respectively. Of the 510 lipids found by manual inspection, XCMS Online, MZmine, and MS-DIAL recovered 461, 429, and 456 respectively. Out of the 510 lipids investigated, 101 were defined as low abundant with a peak intensity of ≤ 5,000 peak height units. Of these low abundant peaks, XCMS Online recovered 85 compared with MZmine and MS-DIAL which recovered 67. XCMS Online generated a peak list that contained 70.3% false positives, whereas the peak list from MZmine and MS-DIAL contained 29.8% and 23.2% respectively. The ability to deconvolute isomers revealed MS-DIAL with the superior peak picking algorithm with a detection of 51 out of 53 isomers, whereas XCMS Online and MZmine and recovered 42 and 33 respectively. Peak picking algorithms were further investigated by reviewing the deconvolution of the Triacylglycerol (54:5), Triacylglycerol (49:1), and Phosphatidylcholine (36:3) isomers. Various MZmine deconvolution parameters were tested for Baseline Correction, Noise Amplitude, Local Minimum Search, and Savitzky-Golay algorithms with none displaying the ability to separate peaks of various peak shapes and intensities found in one chromatographic run.