Proteome encoded determinants of protein sorting into extracellular vesicles

Area of Science:

• Biochemistry
• Proteomics
• Bioinformatics

Background:

• Extracellular vesicles (EVs) mediate cell-to-cell communication, and their protein cargo holds biomarker potential.
• Mechanisms governing protein sorting into EVs are not fully understood.
• Accurate identification of EV-associated proteins is crucial for biomarker development.

Purpose of the Study:

• To determine if protein sequence alone can predict EV association.
• To identify key protein features driving EV association.
• To develop an explainable AI model for EV protein prediction.

Main Methods:

• Trained and validated explainable AI models on human proteome data from EV databases.
• Corrected datasets for contaminants and experimental biases (e.g., mass spectrometry).
• Analyzed protein sequence features and post-translational modification (PTM) annotations.

Main Results:

• A sequence-based model achieved an AUC of 0.77 ± 0.01 for predicting EV proteins.
• Incorporating PTM annotations improved prediction accuracy to 0.84 ± 0.00.
• EV-associated proteins are characterized by stability, polarity, structure, and low isoelectric point.
• Palmitoylation emerged as a key PTM for EV sorting, particularly in strictly isolated EVs.

Conclusions:

• Protein sequence analysis, enhanced by PTM data, can effectively predict EV association.
• PTMs, especially palmitoylation, are critical determinants of protein sorting into EVs.
• This AI-driven approach provides a valuable tool for EV proteome characterization and biomarker discovery.