PanoMito: mapping mitochondrial heterogeneity from morphofunctional profiling to single-mitochondrion spatial transcriptomics

<title>Abstract</title> <p>The complexity of the relationship between mitochondrial morphology and function has been widely recognized, yet systematic investigation of this relationship has been hindered by the lack of accurate segmentation tools and comprehensive benchmarks. Here we present PanoMito, an integrated toolbox built upon PanoMitoAtlas, to our knowledge the largest multi-modal fluorescence mitochondrial dataset comprising &gt;130,000 annotated instances across 16 species; and PanoMitoNet, a deep-learning framework unifying instance segmentation and unsupervised classification. PanoMitoNet achieves superior precision (AP₅₀ of 65.0%) compared to existing methods. At single-mitochondrion resolution, PanoMito reveals extensive mitochondrial heterogeneity across species and cell types, providing a quantitative approach to dissect the complex morphology-function relationship without relying on morphological inference alone. Furthermore, integration with in situ sequencing enables single-mitochondrion spatial transcriptomics. By providing single-organelle resolution across morphology, function, and transcriptomics, PanoMito establishes a comprehensive, universally applicable framework for dissecting mitochondrial biology from model systems to non-model organisms.</p>