Liquid chromatography-mass spectrometry (LC-MS)-based lipidomics generates large datasets that need to be interpreted using high-performance data pre-processing tools such as XCMS, mzMine, and Progenesis. These pre-processing tools rely heavily on accurate peak detection, which depends on proper setting of the peak detection mass tolerance (PDMT). The PDMT is usually set with a fixed value in either ppm or Da units. However, this fixed value may result in duplicates or missed peak detection and inaccurate peak quantification. To improve the accuracy of peak detection, we developed the dynamic binning method, which considers peak broadening described by the physics of ion separation and sets the PDMT dynamically in function of m/z. In our method, the PDMT is proportional to (mz)2 for Fourier-transform ion cyclotron resonance (FTICR), to (mz)1.5 for Orbitrap and to m/z for Quadrupole time-of-flight (Q-TOF), and is a constant for Quadrupole mass analyzer. The dynamic binning method was implemented in XCMS [1,2], and the adopted source code is available in GitHub at https://github.com/xiaodfeng/DynamicXCMS. We have compared the performance of the XCMS implemented dynamic binning with different popular lipidomics pre-processing tools to find differential compounds. We generated set samples with 43 lipid internal standards that were differentially spiked to aliquots of one human plasma lipid sample using Orbitrap LC-MS/MS. The performance of various pipelines using matched parameter sets was quantified by a quality score system that reflects the ability of a pre-processing pipeline to detect differential peaks spiked at various concentrations. The quality score indicated that our dynamic binning method improves the quantification performance of XCMS (maximum p-value 9.8·10−3 of two-sample Wilcoxon test) over its original implementation. We also showed that the XCMS with dynamic binning found differential spiked-in lipids better or with similar performance as mzMine and Progenesis do.
Read full abstract