Contrastive Learning

Nano-Org's Contrastive Learning tools enable users to compare nanoscale protein organisation across multiple single-molecule localisation microscopy (SMLM) datasets using machine-learning techniques developed by the Owen Research Group at the University of Birmingham. Rather than relying on predefined measurements, contrastive learning extracts representations directly from localisation data. These representations, known as embeddings, can be used to identify structurally similar nanoscale architectures and explore relationships between proteins, cell types, experimental conditions, and biological systems.

Who Can Use Contrastive Learning?

Contrastive Learning is available to registered Nano-Org users.

What Is an Embedding?

An embedding is a numerical representation of a dataset learned directly from its spatial organisation. Datasets with similar nanoscale structures are positioned closer together in the embedding space, while structurally distinct datasets are positioned further apart.

This allows relationships between datasets to be explored in a way that may not be apparent from metadata alone.

Submitting an Analysis

To perform a contrastive learning analysis:

  1. Navigate to the Contrastive Learning page.
  2. Select one or more datasets from the available dataset table.
  3. Review your selected datasets.
  4. Submit the analysis job.
  5. Download the generated outputs once processing has completed.

Analyses are performed using University of Birmingham computational resources and may take several minutes depending on the number and size of the selected datasets. You should receive an email notification when your analysis is complete.

Example Applications

  • Comparing protein organisation across different cell types.
  • Investigating the effects of treatments or experimental perturbations.
  • Identifying biologically related nanoscale phenotypes.
  • Exploring relationships between independent studies.
  • Generating features for downstream machine-learning analyses.
  • Supporting spatial nano-omics investigations.

Associated Publication

If you use Contrastive Learning analyses generated by Nano-Org in your research, please cite:

Nanoscale spatial-omics via contrastive embedding of single-molecule localisation data. S. Shirgill, L.G. Jensen, D.J. Nieves, M.C. Wales, A. Kaminer, K. Savoye, R. Peters, M. Heilemann, R. Henriques, S.F. Lee, P. Rubin-Delanchy and D.M. Owen. bioRxiv (2025). DOI: 10.1101/2025.10.07.679170