metaboprep: an R package for preanalysis data description and processing

Abstract

Motivation: Metabolomics is an increasingly common part of health research and there is need for preanalytical data processing. Researchers typically need to characterize the data and to exclude errors within the context of the intended analysis. Whilst some preprocessing steps are common, there is currently a lack of standardization and reporting transparency for these procedures.

Results: Here, we introduce metaboprep, a standardized data processing workflow to extract and characterize high quality metabolomics datasets. The package extracts data from preformed worksheets, provides summary statistics and enables the user to select samples and metabolites for their analysis based on a set of quality metrics. A report summarizing quality metrics and the influence of available batch variables on the data are generated for the purpose of open disclosure. Where possible, we provide users flexibility in defining their own selection thresholds.

Availability and implementation: metaboprep is an open-source R package available at https://github.com/MRCIEU/metaboprep.

Supplementary information: Supplementary data are available at Bioinformatics online.

Fig. 1.

Brief description of the metaboprep pipeline. Along the top, the six primary steps the pipeline takes are outlined. The column on left provides an outline of the steps for the generation of summary statistics whilst the right provides an outline of the steps taken for sample and metabolite filtering. Common abbreviations used are: ‘dme’ for derived measures excluded; SD for standard deviations; ‘X’ which denotes a threshold variable that is defined by the user in the pipeline parameter file; PCs for principal components

Fig. 2.

Summary figure found in each HTML report for the filtered dataset. There are seven figures in this BiB dataset summary figure. (1) The distribution of sample missingness. (2) The distribution for feature missingness. (3) The distribution for TSA, at complete features only. (4) A hierarchical clustering dendrogram based on absolute Spearman rho distances (1-rho) and cut at a tree cut height (red horizontal line) defined by the user. Blue branches on the dendrogram denote the features specified as ‘representative’ features used in the PCA. (5) A table of the number of metabolites used at each step of the dendrogram and PCA. (6) A scree plot of the variance explained for each PC also identifying the number PCs estimated to be informative (vertical lines) by the Cattel’s Scree Test acceleration factor (red, n = 2) and Parallel Analysis (green, n = 49). (7) A PC plot of the top two PCs for each sample. The number of metabolites used in the analysis is again indicated in the title of the PC plot. Individuals in the PC plot were clustered into four k-means (k) clusters, using data from the top two PCs. The k-means clustering and colour coding is strictly there to help provide some visualization of the major axes of variation in the sample population(s)