Determinants of cysteine coverage by chemoproteomics

Covalent chemoproteomics has emerged as a potent method to broaden the druggable proteome by screening compounds that covalently modify nucleophilic residues, primarily cysteine. However, conventional experiments only uncover 5-18% of the approximately 250,000 cysteines encoded by the human genome, leaving the factors contributing to this limited coverage largely unknown.

This study combines cysteine feature generation, explainable machine learning (ML), and robust statistical filtering to determine what influences cysteine detectability in iodoacetamide-desthiobiotin (IA-DTB) chemoproteomics. Employing a scalable feature-generation pipeline, the researchers calculated physicochemical and structural properties for cysteines and their corresponding tryptic peptides, subsequently training models to predict detectability with high accuracy.

Shapley analysis reveals that protein abundance and intrinsic peptide properties associated with LC-MS behavior are the primary factors governing detection, while cysteine structural descriptors contribute minimally, suggesting that detection is limited after probe reactivity. Additionally, the study introduces a distance-based statistical filter that enhances reproducibility across replicates.

Furthermore, the researchers demonstrate that selecting specific cell lines, performing subcellular fractionation, and utilizing alternative electrophiles can collectively expand coverage by up to 30%. Overall, this work shifts the optimization of chemoproteomics workflows from a trial-and-error approach to a data-driven, rational design.