Deep Learning Enhances Precision of Citrullination Identification in Human and Arabidopsis Tissue Proteomes

Citrullination is a key yet underexplored post-translational modification involved in various biological processes. Its identification via mass spectrometry faces challenges like limited enrichment tools and false positives due to mass overlap with deamidation (+0.9840 Da). To address this, we developed a data analysis pipeline integrating the deep learning model Prosit-Cit, trained on ~53,000 spectra from ~2,500 synthetic citrullinated peptides, which improves sensitivity and precision in identifying citrullination sites. This approach has identified up to 14 times more citrullinated sites in human tissue proteomes and revealed new insights, including the first large-scale citrullination mapping in Arabidopsis. This upload includes: 1) Raw files and search SEARCHs from the evaluation dataset, used to assess the precision of citrullination identifications. 2) Raw files, search SEARCHs, and rescoring outcomes from validation experiments conducted on Arabidopsis flowers. 4) Re-analyzed search and rescoring SEARCHs from human (PXD010154) and Arabidopsis (PXD013868) tissue proteomes.