MSlineaR: An R Package Assessing Linear Behavior to Improve Quality Assurance and Statistical Robustness in Untargeted Metabolomics

Abstract Mass spectrometry-based untargeted metabolomics analyzes complex biological matrices containing thousands of individual features. Linearity is a key analytical parameter in quantitative mass spectrometry, reflecting proportionality between signal intensity and analyte concentration within a defined range. In untargeted metabolomics, linearity cannot be directly assessed due to the absence of reference concentrations. Instead, the range in which features exhibit approximately linear dilution-dependent behavior (ALB) can be evaluated as a practical proxy. Feature selection based on this response enables the early removal of noise and unreliable features, thereby reducing the risk of false-positive findings and improving analytical robustness, while the reduced number of retained features lowers the multiple-testing burden in downstream statistical analyses. We present MSlineaR, an open-source R-based software tool implementing a six-step process to assess dilution-dependent response behavior in metabolomic data sets. MSlineaR evaluates dilution curves to identify nonclassical response patterns, detect outliers, and iteratively trim boundary regions to remove plateau effects. It then reassesses the remaining data to retain features exhibiting ALB. Importantly, it defines boundaries of the approximately linear range (ALR) and applies them for data curation, enabling exclusion of unreliable signals while preserving robust features. Application to three independent data sets demonstrated a ∼10% improvement in the number of features classified as exhibiting ALB compared to classical linear regression-based approaches. MSlineaR was complementary to relative standard deviation (RSD) filtering, improved median RSD values and enhanced the robustness of statistical modeling.