Contextual AI models for single-cell protein biology.

Area of Science:

• Computational biology
• Genomics
• Molecular biology

Background:

• Understanding protein function and interactions is crucial for molecular therapies.
• Existing algorithms struggle to model protein interactions across diverse biological contexts.
• Deciphering cell types and protein roles is essential for biological research.

Purpose of the Study:

• Introduce PINNACLE, a geometric deep learning approach for context-aware protein representations.
• Generate comprehensive protein representations across various cell types and tissues.
• Enhance the modeling of protein interactions within specific biological contexts.

Main Methods:

• Leveraged a multiorgan single-cell atlas for training.
• Employed geometric deep learning to learn contextualized protein interaction networks.
• Generated 394,760 protein representations across 156 cell type contexts in 24 tissues.

Main Results:

• PINNACLE's embedding space captured cellular and tissue organization, enabling zero-shot retrieval of tissue hierarchy.
• Pretrained representations improved 3D structure-based immuno-oncological interaction analysis and drug effect investigation.
• Outperformed state-of-the-art models in identifying therapeutic targets for rheumatoid arthritis and inflammatory bowel diseases.

Conclusions:

• PINNACLE provides context-specific protein representations, advancing biological predictions.
• The method enhances the development of targeted molecular therapies and drug discovery.
• Enables large-scale, context-specific biological predictions for future research.