Reproduce “A vision foundation model for single-cell biology via spatial gene cartography” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Ridvan Yesiloglu, Sakib Mostafa, James Zou, Ash Alizadeh, Jiajun Wu, Lei Xing, Ehsan Adeli, Md Tauhidul Islam

Researchers at Stanford University developed `scVision`, a vision foundation model for single-cell biology that represents transcriptomes as spatially organized images to capture continuous gene expression and biological relationships. This model demonstrated superior zero-shot cell-type annotation across diverse datasets and achieved up to 300-fold higher encoding throughput than token-based models.

Key insights

  • The spatial organization of genes in `scImage` is critical, significantly improving annotation accuracy compared to random gene placement, indicating the learned layout's importance for biological signal.
  • `scVision`'s attention maps provide biologically interpretable insights, revealing cell-type-specific and recurrent gene programs across diverse tissues without explicit pathway supervision.
  • The fixed-size image representation enables significantly higher computational efficiency and strong robustness to gene dropout, maintaining performance even with substantial data corruption.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.