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

Area of Science:

• Proteomics
• Post-translational Modifications
• Bioinformatics

Background:

• Citrullination is a vital post-translational modification (PTM) crucial for biological processes but remains understudied.
• Mass spectrometry struggles with accurate citrullination site identification due to challenges like false positives and limited enrichment tools.

Purpose of the Study:

• To develop a novel data analysis pipeline for sensitive and precise proteome-wide identification of citrullination sites.
• To overcome limitations in current mass spectrometry-based approaches for studying citrullination.

Main Methods:

• Integration of the deep learning model Prosit-Cit into a mass spectrometry (MS) database search workflow.
• Prosit-Cit trained on synthetic citrullinated peptides to predict retention time and fragment ion intensities.
• Application of the pipeline to human tissue proteomes and the plant *Arabidopsis thaliana*.

Main Results:

• The pipeline demonstrated high precision, recovering most known citrullination sites in human tissues.
• Achieved up to a 14-fold increase in sensitivity for identifying citrullinated sites.
• Identified approximately 200 citrullination sites in *Arabidopsis thaliana*, marking the first large-scale mapping in plants.

Conclusions:

• The developed pipeline significantly enhances the accuracy and throughput of citrullination site identification.
• This robust tool facilitates deeper understanding of protein citrullination's role in health, disease, and across diverse species.
• Enables seamless application to existing proteomics data for accelerated biological discovery.