A0668
Title: A conformal prediction framework for outlier detection in mass spectrometry
Authors: Soohyun Ahn - Ajou University (Korea, South) [presenting]
Abstract: Ensuring the reliability of mass spectrometry (MS) data is essential for biomarker discovery and the analysis of complex biological systems. However, existing outlier detection methods are often limited by their focus on either sample- or peak-level outliers, reliance on subjective criteria, and use of fixed thresholds derived from asymptotic distributions. These limitations hinder their ability to fully capture the variability in MS data. CPOD (Conformal Prediction for Outlier Detection) is introduced, a novel method that applies conformal prediction to detect both outlier samples and outlier peaks in a unified, data-driven, and distribution-free framework. Through extensive numerical evaluations, CPOD demonstrates superior performance compared to existing approaches. Additionally, its application to real LC-MRM data illustrates its effectiveness in improving data quality and reproducibility, making it a promising tool for MS-based studies.
