Unifying proteomic technologies with ProteinProjector

Area of Science:

• Proteomics
• Systems Biology
• Bioinformatics

Background:

• Subcellular protein organization is crucial for cellular function.
• Existing proteomic methods offer limited scope and sensitivity for mapping protein localization.
• Integrating diverse data types is challenging but necessary for comprehensive understanding.

Purpose of the Study:

• To develop a novel deep learning framework, ProteinProjector, for integrating multi-modal proteomic data.
• To create a unified map of protein subcellular positions by leveraging diverse datasets.
• To enhance the accuracy and coverage of subcellular localization predictions.

Main Methods:

• Developed a self-supervised deep learning framework named ProteinProjector.
• Integrated four proteome-wide datasets from HEK293 cells: AP-MS, PL-MS, SEC-MS, and fluorescent imaging.
• Evaluated the framework's performance against individual modalities and other integration methods.

Main Results:

• ProteinProjector successfully integrates diverse proteomic data for unified protein mapping.
• Map coverage and accuracy significantly improved with the addition of more data modalities.
• Maximal recovery of known protein complexes was achieved using all four integrated datasets.
• ProteinProjector outperformed individual methods in predicting orthogonal functional and physical associations.

Conclusions:

• ProteinProjector provides a robust foundation for integrating diverse data modalities to characterize subcellular structure.
• The framework offers improved accuracy and coverage for mapping protein localization compared to existing methods.
• This approach facilitates a more comprehensive understanding of cellular organization and protein function.