A Unified Framework for Batch Correction and Imputation in Modern Mass Spectrometry Proteomics

Abstract

Mass spectrometry-based proteomics, including single-cell applications, are often confounded by batch effects, inter-laboratory variation, and signal drift that can obscure biological interpretation. NMFBatch addresses these sources of unwanted variation through a unified framework that combines non-negative matrix factorization with generalized additive modelling to jointly capture discrete and continuous technical effects from protein expression data. It handles missing values directly through masking and on-the-fly imputation. Benchmarking across reference datasets, plasma proteomics cohorts, simulations, and single-cell data demonstrated that NMFBatch reduced technical variation while preserving biological structure. These results establish NMFBatch as a robust framework for batch correction, harmonization, and integrative proteomics analysis from protein expression data.