MitoAtlas provides domain-resolution spatial mapping of 861 mitochondrial proteins through multiplexed super-resolution proximity labeling integrated with machine learning classification, resolving membrane topologies for 196 transmembrane proteins, identifying 160 previously unannotated mitochondrial proteins, and enabling prediction of protein-protein and protein-metabolite interactions that reveal compartment-specific organizational principles of mitochondrial architecture and metabolism.

Mitochondrial function depends on the precise sub-organellar localization and topological orientation of its proteome. However, mapping this architecture with domain-level precision across the double-membrane system has remained challenging. Here, we present MitoAtlas, a spatial systems biology resource built on SR-PL (super-resolution proximity labeling) that resolves the sub-mitochondrial architecture of 861 proteins at domain-level resolution. By deploying 42 spatially restricted APEX2 and BioID/BioID2 baits across mitochondrial compartments and cytosol in HEK293T-REx cells, we mapped 13,440 unique proximity-labeled sites. Integrating high-density spatial data with a machine-learning classification pipeline and AlphaFold structural modeling, we resolved membrane topologies for 196 transmembrane proteins and assigned sub-mitochondrial localizations to 160 previously unannotated mitochondrial proteins absent from MitoCarta3.0, including MINPP1 as an intermembrane space resident. Coupling bait-specific proximity labeling signatures with AlphaFold structure prediction and metabolite docking identified 47 protein–protein interactions, including the LETM1–Citrate Synthase complex, and 155 protein–metabolite interactions. MitoAtlas is available at mitoatlas.org as a domain-resolution spatial map of mitochondrial architecture and metabolism.