description.md

Universal Cell Embeddings - Demo Instructions

Overview

This demo allows users to upload a .h5ad file and process it using a pre-trained Universal Cell Embedding (UCE) model. The demo has two primary functionalities:

1. Generate Embeddings: Upload a .h5ad file, and the model will output the embeddings for the loaded cells.
2. Cell-Type Classification: For cells from the Tabula Sapiens v1 dataset, users can further classify the cell types based on the embeddings.

Features

• Upload and Embedding Generation: Upload your .h5ad file containing single-cell RNA sequencing data, and the UCE model will generate embeddings representing the cells.
• Human Cell-Type Classification: Specifically for human data from the Tabula Sapiens v1 dataset, the demo can classify cells into known cell types based on the generated embeddings.

How to Use the Demo

1. Upload a .h5ad File:

   • Click on the "Upload File" button.
   • Select your .h5ad file from your device. Ensure that the file is formatted correctly.

2. Generate Embeddings:

   • After the file is uploaded, click on the "Generate Embeddings" button.
   • The model will process the data and output the embeddings for each cell.

3. Cell-Type Classification (Optional):

   • If your data comes from the Tabula Sapiens v1 dataset, you can click on the "Classify Cell Types" button after generating embeddings.
   • The demo will output the predicted cell types based on the embeddings.

Supported Data

• Input Format: The demo accepts .h5ad files.
• Supported Dataset: The classification functionality is tailored for human data from Tabula Sapiens v1. Other datasets will only generate embeddings without classification.

Example Workflow

1. Step 1: Upload your .h5ad file.
2. Step 2: Click "Generate Embeddings" to get the cell embeddings.
3. Step 3: (Optional) If working with Tabula Sapiens v1, click "Classify Cell Types" for classification results.

Future Enhancements

• Expanding the classification functionality to support more datasets and species.
• Providing visualization tools for the embeddings directly within the demo.

Limitations

• The classification feature is limited to Tabula Sapiens v1 dataset.
• For other datasets, only the embeddings will be generated without classification.

Contact and Feedback

For any issues or feedback, please contact the development team through the Hugging Face community or open an issue in the repository.