NanTex enables computational multiplexing and phenotyping of organelles across super-resolution modalities

Super-resolution microscopy (SRM) enables nanoscale visualization of cellular organelles but remains constrained by spectral overlap, labeling requirements, and temporal offsets in multicolor imaging. We present NanTex , a deep learning framework that introduces nanotexture as a universal descriptor of subcellular organization and achieves probabilistic demixing of multiple organelles from single-channel SRM images. Unlike segmentation methods that enforce exclusivity, NanTex preserves overlapping morphologies, enabling faithful reconstruction even in crowded regions. Trained on small curated datasets with augmentation, NanTex generalized across SMLM, MINFLUX, STED, SIM, and live-cell Airyscan, with modality-specific retraining where necessary. Demonstrated on cytoskeletal, endomembrane, and metabolic organelles, NanTex achieved high-fidelity reconstructions and extended uniquely to live-cell imaging, where it eliminated temporal misalignment and enabled quantitative tracking of vesicle-like ER subdomains. Notably, NanTex enables quantitative dynamic readouts directly from single-channel live-cell data, revealing and tracking hidden subdomains in real time without the need for additional labels. Beyond multiplexing, NanTex supported computational phenotyping by distinguishing structurally distinct yet molecularly identical populations, exemplified by nocodazole-induced microtubule depolymerization and its glyoxal-mediated modulation. Together, these results establish nanotexture as a paradigm-shifting descriptor for organelle identity and position NanTex as a modality-agnostic, label-efficient, and live-cell-ready strategy for quantitative nanoscale biology.