GraphPI: Efficient Protein Inference with Graph Neural Networks.

Area of Science:

• Biomedical research
• Computational biology
• Proteomics

Background:

• Deep learning has advanced biomedical research but faces challenges in protein inference due to scarce labeled data and high annotation costs.
• Accurate protein inference is crucial for understanding biological processes and disease mechanisms.

Purpose of the Study:

• To introduce GraphPI, a novel graph neural network-based framework for protein inference.
• To address the challenge of limited labeled data in protein inference using pseudolabels and self-training.
• To develop a universally applicable protein inference method that avoids dataset-specific fine-tuning.

Main Methods:

• GraphPI models protein inference as a node classification problem within a protein-peptide-PSM graph.
• The framework utilizes a graph neural network architecture to analyze interrelations between proteins.
• Pseudolabels from existing algorithms and self-training refine model training on unlabeled public datasets.

Main Results:

• GraphPI demonstrates universal applicability across different datasets without requiring dataset-specific fine-tuning, mitigating overfitting.
• The model achieves notable performance on various test datasets.
• GraphPI significantly reduces computation times compared to conventional protein inference algorithms.

Conclusions:

• GraphPI offers an efficient and accurate solution for protein inference, overcoming data scarcity limitations.
• The framework's universal applicability and computational efficiency make it a valuable tool for proteomics research.
• This approach advances the application of deep learning in large-scale protein identification.