Grape-Pi: graph-based neural networks for enhanced protein identification in proteomics pipelines

Area of Science:

• Proteomics
• Bioinformatics
• Computational Biology

Background:

• Mass spectrometry (MS) is key for untargeted protein identification but faces challenges with data complexity and false discovery rates (FDR).
• Accurate protein identification is crucial for understanding biological processes and disease mechanisms.

Purpose of the Study:

• To develop an advanced computational model for improving protein identification accuracy in mass spectrometry.
• To leverage protein-protein interaction (PPI) data to enhance the performance of proteomics pipelines.

Main Methods:

• Developed a graph neural network (GNN)-based model named Grape-Pi (Graph Neural Network using Protein-Protein Interaction for Enhancing Protein Identification).
• Integrated PPI data using two types of message-passing layers to incorporate evidence from target proteins and their interactors.
• Applied the model to various proteomics datasets, including yeast and gastric samples.

Main Results:

• Grape-Pi significantly improved the area under the receiver-operating characteristic curve (AUC) for protein identification, showing 18% and 7% gains in yeast samples and 9% in gastric samples compared to traditional methods.
• Proteins identified by Grape-Pi in gastric samples correlated highly with mRNA data.
• The model successfully identified key gastric cancer proteins, such as MAP4K4, that were missed by conventional techniques.

Conclusions:

• Grape-Pi offers a powerful new approach to enhance protein identification accuracy in mass spectrometry-based proteomics.
• The model's ability to integrate PPI data provides a more comprehensive analysis, leading to improved detection of biologically relevant proteins.
• Grape-Pi is applicable across diverse proteomics pipelines and demonstrates potential for biomarker discovery in diseases like cancer.