Data-adaptive pipeline for filtering and normalizing metabolomics data

Introduction Untargeted metabolomics datasets contain large proportions of uninformative features and are affected by a variety of nuisance technical effects that can bias subsequent statistical analyses. Thus, there is a need for versatile and data-adaptive methods for filtering and normalizing data prior to investigating the underlying biological phenomena. Objectives Here, we propose and evaluate a data-adaptive pipeline for metabolomics data that are generated by liquid chromatography-mass spectrometry platforms. Methods Our data-adaptive pipeline includes novel methods for filtering features based on blank samples, proportions of missing values, and estimated intra-class correlation coefficients. It also incorporates a variant of k-nearest-neighbor imputation of missing values. Finally, we adapted an RNA-Seq approach and R package, scone , to select an appropriate normalization scheme for removing unwanted variation from metabolomics datasets. Results Using two metabolomics datasets that were generated in our laboratory from samples of human blood serum and neonatal blood spots, we compared our data-adaptive pipeline with a traditional filtering and normalization scheme. The data-adaptive approach outperformed the traditional pipeline in almost all metrics related to removal of unwanted variation and maintenance of biologically relevant signatures. The R code for running the data-adaptive pipeline is provided with an example dataset at . Conclusion Our proposed data-adaptive pipeline is intuitive and effectively reduces technical noise from untargeted metabolomics datasets. It is particularly relevant for interrogation of biological phenomena in data derived from complex matrices associated with biospecimens.