MetaboLyzer: A Novel Statistical Workflow for Analyzing Postprocessed LC–MS Metabolomics Data

Abstract

Metabolomics, the global study of small molecules in a particular system, has in the last few years risen to become a primary –omics platform for the study of metabolic processes. With the ever-increasing pool of quantitative data yielded from metabolomic research, specialized methods and tools with which to analyze and extract meaningful conclusions from these data are becoming more and more crucial. Furthermore, the depth of knowledge and expertise required to undertake a metabolomics oriented study is a daunting obstacle to investigators new to the field. As such, we have created a new statistical analysis workflow, MetaboLyzer, which aims to both simplify analysis for investigators new to metabolomics, as well as provide experienced investigators the flexibility to conduct sophisticated analysis. MetaboLyzer’s workflow is specifically tailored to the unique characteristics and idiosyncrasies of postprocessed liquid chromatography/mass spectrometry (LC/MS) based metabolomic datasets. It utilizes a wide gamut of statistical tests, procedures, and methodologies that belong to classical biostatistics, as well as several novel statistical techniques that we have developed specifically for metabolomics data. Furthermore, MetaboLyzer conducts rapid putative ion identification and putative biologically relevant analysis via incorporation of four major small molecule databases: KEGG, HMDB, Lipid Maps, and BioCyc. MetaboLyzer incorporates these aspects into a comprehensive workflow that outputs easy to understand statistically significant and potentially biologically relevant information in the form of heatmaps, volcano plots, 3D visualization plots, correlation maps, and metabolic pathway hit histograms. For demonstration purposes, a urine metabolomics data set from a previously reported radiobiology study in which samples were collected from mice exposed to gamma radiation was analyzed. MetaboLyzer was able to identify 243 statistically significant ions out of a total of 1942. Numerous putative metabolites and pathways were found to be biologically significant from the putative ion identification workflow.

The effect of low dose rate on metabolomic response to radiation in mice

“…For instance, creatinine relative intensity in the control mice over the course of the study was 32.87 ± 1.87, and for the HDR and LDR mice, these values were 34.60 ± 1.39 and 34.20 ± 1.04, respectively. The normalized data were then analyzed using traditional statistical tools such as SIMCA-P + , Random Forests (RF), and Metabolyzer (Mak et al 2014) to determine whether the urinary metabolomic profile of LDR-irradiated mice could be distinguished from that of HDR-irradiated mice at matching doses. Although LDR and HDR share many similarities in their urinary metabolomic profiles, subtle and unique differences were observed with dose rate.…”

“…An in-house statistical algorithm called Metabolyzer (Mak et al 2014) was used along with SIMCA-P+ (Umetrics, Umea, Sweden) and Random Forests (RF) to determine the metabolites that contributed the most to the differentiation between urinary profiles of irradiated samples and that of the controls in LDR and HDR groups. When comparing data from control versus radiationexposed animals, metabolites with non-zero abundance values in at least 70 % of samples in both groups (com-plete-presence ions) were first identified.…”

Metabolic Dysregulation after Neutron Exposures Expected from an Improvised Nuclear Device

“…Univariate statistical analysis was performed on the neutron vs. X ray groups (1 Gy) at days 1 and 7 postirradiation with the in-house statistical software, MetaboLyzer (30). For complete presence ions (≥75% presence in each group), statistically significant ions ( P < 0.05) were identified with Welch’s t test with a false discovery rate (FDR) of 0.2, while statistically significant ( P < 0.05) partial presence ions (<75% presence in each group) with FDR = 0.2 were identified with the categorical Barnard’s test.…”

Gas Chromatography/Mass Spectrometry Metabolomics of Urine and Serum from Nonhuman Primates Exposed to Ionizing Radiation: Impacts on the Tricarboxylic Acid Cycle and Protein Metabolism

“…Volcano plots (of analytes present in ≥70% of samples in a group, false discovery rates determined by Benjamini–Hochberg step-up procedure, significant ions determined with Welch’s t test) were generated with the in-house statistical software MetaboLyzer, and unsupervised standard singular value decomposition-based principal component analysis (PCA) plots were generated. 35 Statistically significant ions were determined with a Kruskal–Wallis test and a posthoc Duncan test in SAS 9.4 (Cary, NC). Significant ions were validated by matching their retention indices and EI spectra to authentic standards or the LECO/Fiehn Metabolomics Library and comparison with the NIST mass spectral library.…”