Learning consistent subcellular landmarks to quantify changes in multiplexed protein maps

Abstract

Highly multiplexed quantitative subcellular imaging holds enormous promise for understanding how spatial context shapes the activity of our genome and its products at multiple scales. Yet unbiased analysis of subcellular organisation across experimental conditions remains challenging, because differences in molecular profiles between conditions confound differences in molecular profiles across space. Here, we introduce a deep-learning framework called CAMPA (Conditional Autoencoder for Multiplexed Pixel Analysis), which uses a variational autoencoder conditioned on cellular states and perturbations to learn consistent molecular signatures. Clustering the learned representations into subcellular landmarks allows quantitative comparisons of landmark sizes, shapes, molecular compositions and relative spatial organisation between conditions. By performing high-resolution multiplexed immunofluorescence on human cells, we use CAMPA to reveal how subnuclear organisation changes upon different perturbations of RNA production or processing, and how different membraneless organelles scale with cell size. Furthermore, by integrating information across the cellular and subcellular scales, we uncover new links between the molecular composition of membraneless organelles and bulk RNA synthesis rates of single cells. We anticipate that CAMPA will greatly accelerate the systematic mapping of multiscale atlases of biological organisation to identify the rules by which context shapes physiology and disease.