C-COMPASS places artificial intelligence at the center of spatial omics: The tool developed by Helmholtz Munich, the DZD and the University of Bonn uses neural networks to reliably predict multiple subcellular localizations of proteins. These AI models combine the software with whole-proteome data to quantify shifts in protein distribution and in organelle abundance – a task at which many conventional, rule-based approaches fail.

Why AI design makes the difference:

• AI-based prediction: Neural networks capture complex patterns and multiple localizations of individual proteins instead of 'either-or' assignments.

• Data fusion via AI: proteome and lipidome data are merged in a unified workflow. Lipids are localized via proteomic reference maps supported by AI – a breakthrough for spatial lipidomics.

• Reproducibility and accessibility: Standardized processing steps and a graphical interface bring AI analytics into laboratories without programming knowledge.

„With C-COMPASS we wanted to make the strengths of AI usable for spatial proteomics and make results more reproducible,“ says developer Daniel Haas. Project leader Dr. Natalie Krahmer adds: „For the first time we can systematically address spatial lipidomics by integrating AI-supported proteome and lipidome data and creating cellular atlases on both levels.“

Practical demonstration: In humanized liver tissue, C-COMPASS AI-supported protein distributions and their changes under different metabolic conditions were mapped. By projecting lipid profiles onto proteomic reference maps, the spatial lipid distribution in the mouse model was visualized and correlated with metabolic disorders.

Outlook: The team is working on expanding to additional datasets and disease models as well as on integrating additional spatial omics such as spatial transcriptomics. The goal is to create AI-supported atlases of dynamic cell states and thus identify disease-relevant changes even faster and more precisely.