SubCell: Vision foundation models for microscopy capture single-cell biology

Abstract

Cells are the functional units of life, and the wide range of biological functions they perform are orchestrated by myriad molecular interactions within an intricate subcellular architecture. This cellular organization and functionality can be studied with microscopy at scale, and machine learning has become a powerful tool for interpreting the rich information in these images. Here, we introduce SubCell, a suite of self-supervised deep learning models for fluorescence microscopy that are designed to accurately capture cellular morphology, protein localization, cellular organization, and biological function beyond what humans can readily perceive. These models were trained using the metadata-rich, proteome-wide image collection from the Human Protein Atlas. SubCell outperforms state-of-the-art methods across a variety of tasks relevant to single-cell biology. Remarkably, SubCell generalizes to other fluorescence microscopy datasets without any finetuning, including dataset of drug-perturbed cells, where SubCell accurately predicts drug perturbations of cancer cells and mechanisms of action. Finally, we construct the first proteome-wide hierarchical map of proteome organization that is directly learned from image data. This vision-based multiscale cell map defines cellular subsystems with large protein-complex resolution, reveals proteins with similar functions, and distinguishes dynamic and stable behaviors within cellular compartments. In conclusion, SubCell enables deep image-driven representations of cellular architecture applicable across diverse biological contexts and datasets.