Visualization and normalization of drift effect across batches in metabolome-wide association studies

As a powerful phenotyping technology, metabolomics has provided new opportunities in biomarker discovery through metabolome-wide association studies (MWAS) and identification of metabolites having regulatory effect in various biological processes. While MAS-based metabolomics assays are endowed with high-throughput and sensitivity, large-scale MWAS are doomed to long-term data acquisition generating an overtime-analytical signal drift, inherent to liquid chromatography-MS technique that can hinder the uncovering of true biologically relevant changes. We developed dbnorm, a package in R environment, which allows visualization and removal of signal heterogeneity from large metabolomics datasets. dbnorm integrates advanced statistical tools to inspect dataset structure, at both macroscopic (sample batch) and microscopic (metabolic features) scales. To compare model performance on data correction, dbnorm assigns a score, which allows the straightforward identification of the best fitting model for each dataset. Herein, we show how dbnorm efficiently removes signal drift among batches to capture the true biological heterogeneity of data in two large-scale metabolomics studies.