Conformalized outlier detection for mass spectrometry data

Quality control procedures are crucial for ensuring the reliability of mass spectrometry (MS) data, vital in biomarker discovery and understanding complex biological systems. However, existing methods often concentrate solely on either sample or peak outlier detection, rely on subjective criteria, and employ overly uniform thresholds based on asymptotic distributions, thereby failing to adequately capture the characteristics of the data. In this paper, we introduce a novel approach, CPOD (Conformal Prediction for Outlier Detection), leveraging conformal prediction for outlier detection in MS data analysis. CPOD simultaneously identifies outlier samples and peaks based on data-driven and distribution-free principles. Rigorous numerical evaluations and comparisons with existing methods demonstrate superior diagnostic performance. Application to real LC-MRM data underscores practical utility, enhancing data reliability and reproducibility in MS studies. • Introduces a novel conformal prediction approach for outlier detection in mass spectrometry data. • Utilizes Mahalanobis distance as a conformity score to handle both homogeneous and heterogeneous data. • Simultaneously detects outlier samples and peaks using data-driven principles. • Validates superior diagnostic performance compared to the existing methods. • Application to real LC-MRM data strengthens quality control and enhances data reliability and reproducibility.