A Comprehensive Model for Separating Systematic Bias and Noise in Metabolomic Timecourse Data—A Nonlinear B‐Spline Mixed‐Effects Approach

Abstract

The simultaneous detection of tens to hundreds of metabolites in a single metabolomic timecourse sample offers a unique but often unrealized opportunity for quantification validation. An individual timecourse fit for each metabolite fundamentally convolutes measurement noise with systematic sample bias (stemming from, for example, variable sample dilution, extraction, and normalization). However, since systematic bias, by its definition, influences all metabolites within a sample in a similar fashion, it can be identified and corrected through the simultaneous fit of all detected metabolites in a single timecourse model. This study presents a nonlinear B‐spline mixed‐effects model as a convenient formulation capable of estimating and correcting such bias. The proposed model was successfully applied to real cell culture data and validated using simulated timecourse data perturbed with varying degrees of random noise and systematic bias. The model was able to accurately correct systematic bias of 3%–10% to within 0.5% on average for typical data. An R package for the correction model has been developed to facilitate model adoption and use. The proposed nonlinear B‐spline mixed‐effects formulation is general enough for application to a broad range of research areas beyond just cell culture metabolomics.