Integration of single-cell proteomic datasets through distinctive proteins in cell clusters

Area of Science:

• Proteomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell proteomic datasets are increasing due to advanced technologies like mass spectrometry and antibody-based sequencing.
• Integrating these datasets is essential to mitigate batch effects caused by limited sequencing molecules.
• Existing integration methods for transcriptomic data may not be optimal for low-dimensional proteomic datasets.

Purpose of the Study:

• To introduce SCPRO-HI, a novel algorithm for horizontal integration of antibody-based single-cell proteomic datasets.
• To develop a specialized approach for low-dimensional proteomic data integration that addresses batch effects.
• To provide a method extendable to high-dimensional datasets.

Main Methods:

• SCPRO-HI employs a hierarchical cell anchoring technique for cell matching based on protein similarity.
• A variational auto-encoder model is utilized for correcting batch effects in protein abundances without domain mapping.
• The algorithm includes a method for extension to high-dimensional datasets.

Main Results:

• SCPRO-HI demonstrated superior performance compared to state-of-the-art methods on simulated and real-world datasets.
• The algorithm achieved a 75% higher silhouette score, indicating improved clustering and integration quality.
• SCPRO-HI preserved highly variable proteins (HVPs) 13% better than existing methods.
• Competitive performance was observed on transcriptomic datasets, suggesting broader applicability.

Conclusions:

• SCPRO-HI offers an effective solution for integrating antibody-based single-cell proteomic data, particularly for low-dimensional datasets.
• The algorithm's ability to correct batch effects and preserve data integrity enhances downstream analysis.
• SCPRO-HI shows promise for integrating high-dimensional mass-spectrometry-based proteomic data and transcriptomic data.