Robert Hu

Unsupervised quantitative comparison of subcellular proteome organization in morphologically complex cells

Robert Hu, Pablo Camara Department of Genetics and Institute for Biomedical Informatics, Perelman School of Medicine, University of Pennsylvania, 3700 Hamilton Walk, Philadelphia, PA 19104, USA

Understanding how proteins are spatially organized within cells is essential for decoding cellular function. While large-scale imaging assays based on immunofluorescence and endogenous protein tagging have enabled large-scale profiling of subcellular localization patterns, existing computational methods struggle to generalize across diverse cell types, tissue contexts, and disease states. These limitations are driven in part by a reliance on manual feature engineering, large training datasets, and a limited ability to separate true localization differences from confounding factors like cell morphology and other imaging artifacts. Here we present a novel unsupervised algorithm based on applied metric geometry that addresses these challenges. Specifically, we build upon the Gromov-Wasserstein distance to enable robust comparison of subcellular protein localization patterns while accounting for morphological variation. To support scalability across larger imaging datasets, we incorporate a deep metric learning approach that accelerates computation. We validate our method's ability to cluster known organelle-specific protein localization patterns across morphologically distinct cell lines using immunofluorescence data from the Human Protein Atlas, and apply it to characterize the effect of cholesterol inhibition on lysosome organization in iPSC-derived neurons.