scPROTEIN: a versatile deep graph contrastive learning framework for single-cell proteomics embedding.

Area of Science:

• Proteomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell proteomics offers insights into cellular processes but faces challenges.
• Uncertainty in peptide quantification, missing data, batch effects, and noise impede analysis.
• Existing single-cell transcriptome methods are insufficient for proteomic data.

Purpose of the Study:

• To develop a versatile framework for single-cell proteomics data analysis.
• To address multiple analytical challenges within a unified approach.
• To enhance the accuracy and interpretability of single-cell proteomic data.

Main Methods:

• A multitask heteroscedastic regression model for peptide uncertainty estimation.
• Graph contrastive learning for cell embedding generation.
• Integration of peptide uncertainty and cell embeddings in a unified framework.

Main Results:

• scPROTEIN effectively estimates peptide quantification uncertainty.
• The framework successfully denoises protein data and removes batch effects.
• It generates robust single-cell proteomic-specific embeddings.

Conclusions:

• scPROTEIN provides a comprehensive solution for single-cell proteomics data analysis.
• The framework demonstrates efficiency in cell clustering, batch correction, and cell type annotation.
• scPROTEIN facilitates clinical analysis and exploration of spatially resolved proteomic data.