Contextual AI in Protein Biology: Introducing PINNACLE

The article introduces PINNACLE, a geometric deep learning model designed to create context-aware protein representations. By utilizing single-cell transcriptomic data from 24 tissues and protein interaction networks, PINNACLE generates protein embeddings specific to 156 cell types. This approach enhances predictions related to protein interactions, therapeutic target identification, and disease-specific drug effects. The model outperforms existing methods by incorporating cell type and tissue context, enabling zero-shot tissue hierarchy retrieval and improved accuracy in therapeutic applications, including rheumatoid arthritis and inflammatory bowel disease.