Unlock Cross-Species Insights with Universal Cell Embeddings (UCE) on Vecura

What is Universal Cell Embeddings (UCE)?

Universal Cell Embeddings (UCE) is a transformer-based foundation model designed to generate species-agnostic, fixed-dimensional embeddings from single-cell RNA-seq datasets. By grounding gene tokens in protein-sequence space (using ESM-2), UCE overcomes the traditional challenge of gene-vocabulary mismatches across different organisms. It enables researchers to map single-cell expression profiles into a unified 1280-dimensional representation space. This model is exceptionally useful for tasks such as cross-species integration, zero-shot cell-type annotation, and label transfer without requiring task-specific fine-tuning.

What can users do with UCE on Vecura?

With UCE on Vecura, users can:

• Transform diverse datasets: Easily process AnnData files from nine supported species into universal cell embeddings.
• Perform cross-species analysis: Compare cell states across human, mouse, zebrafish, and other organisms within a single shared embedding space.
• Streamline downstream workflows: Generate embeddings that integrate seamlessly into standard analysis pipelines like Scanpy for clustering and UMAP visualization.
• Conduct zero-shot annotation: Utilize the foundation model to perform label transfers without needing custom training on labeled data.

What the output means

The output provides an enriched AnnData file containing the UCE embeddings in the .obsm["X_uce"] field, alongside a raw embedding matrix suitable for immediate numerical analysis. This allows researchers to perform advanced dimensionality reduction, clustering, and cell-type classification.

This output should be used to support scientific decision making. It does not replace experimental validation.

Why this matters

The biological research community generates vast amounts of single-cell data across diverse species, but these datasets are often siloed due to the difficulty of aligning genes and transcriptomic profiles between different organisms. UCE provides a breakthrough by establishing a common "language" for cell states, moving beyond simple gene-name matching to a biologically grounded representation.

By democratizing access to this foundation model, Vecura allows bioinformaticians and biologists to leverage state-of-the-art transformer architecture for their own data. This accelerates cross-species biological discovery and allows for more robust comparative analyses without the need for intensive local infrastructure or complex custom model training.

• Developed by: Snap-Stanford (Rosen et al.)
• Source: Official GitHub Repository
• Reference: Rosen et al., "Universal Cell Embeddings", bioRxiv 2023