Comparative Proteomics Using Machine Learning

Abstract

Comparative proteomics quantifies protein abundance differences across biological conditions to identify disease biomarkers, drug targets, and regulatory mechanisms, and machine learning models that enhance peptide identification, improve quantification accuracy, and extract biological patterns from high-dimensional proteomic data now complement traditional statistical approaches, yet adoption varies across ML application categories. We evaluated 214 ML-based comparative proteomics programmes across centres in France, Switzerland, and Sweden between 2018 and 2022, spanning five application categories: ML-enhanced peptide spectrum matching, data-independent acquisition deconvolution, missing value imputation and batch correction, differential expression and biomarker discovery, and protein complex and pathway inference. A Comparative Proteomics ML Index (CPMLI) was constructed from five sub-scores -- identification sensitivity improvement, quantification accuracy, cross-study reproducibility, biological discovery yield, and computational scalability -- with weights from regression against sustained adoption into proteomics workflows. CPMLI correlated with adoption at r = +0.84 and discriminated adopted from non-adopted methods with an AUC of 0.882. ML-enhanced spectrum matching scored highest (mean CPMLI 0.824), while protein complex inference trailed at 0.598. Only 35.5 percent exceeded the 0.75 threshold. Identification sensitivity carried the largest weight (beta = +0.278), followed by quantification accuracy (beta = +0.230).