Nanoscale spatial-omics via contrastive embedding of single-molecule localisation data

Abstract

Omics approaches have revolutionised biology, and cells can now be routinely characterised on the genomic, transcriptomic and proteomic levels. However, there is an additional pillar; the (nanoscale) spatial organisation of molecules in the cell - information now accessible through super-resolution microscopy. We present a contrastive learning framework for nanoscale spatial-omics that embeds single-molecule localisation microscopy data into a latent space representing protein architecture directly to enabling comparative analysis. Using simulated and experimental data, we demonstrate its ability to enable new bioanalysis capabilities including assessing changes to cellular nanoscale architecture arising from pharmacological treatments, cell type, fluorophore selection or data-processing workflows. The approach supports downstream tasks such as clustering proteins by nanoscale organisation, mapping dose-response trajectories and identifying batch effects in replicate datasets, establishing contrastive learning as a scalable foundation for nanoscale spatial-omics and providing a platform for comparative phenotyping, quality control, and hypothesis generation.