4D spatiotemporal landscape of mitochondrial phenotypes across cellular states unlocked through representation learning

Summary

Researchers introduce MitoSpace, a self-supervised AI model trained on 4D lattice light-sheet microscopy data that can identify drug-induced mitochondrial phenotypes and predict membrane potential from morphology without requiring labeled data. The model demonstrates generalization to unseen perturbations and lung organoids, with improved performance scaling from 2D to 4D imaging.