CytoVerse: Single-Cell AI Foundation Models in the Browser.

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Single-cell RNA sequencing (scRNA-seq) data analysis is crucial for understanding cellular heterogeneity.
• Mapping scRNA-seq datasets to large atlases is often limited by server infrastructure and data privacy concerns.
• Existing methods require significant computational resources and centralized data storage.

Purpose of the Study:

• To introduce CytoVerse, a novel framework for running scRNA-seq Foundation Models (scFM) entirely within a web browser.
• To overcome server constraints and privacy issues associated with large-scale single-cell data analysis.
• To enable scalable and privacy-preserving distributed single-cell analysis.

Main Methods:

• Deployment of scFM models using ONNX for serverless client-side computation.
• Implementation of compressed indexing (IVFPQ) for efficient searching of large cell references (>20 million cells) directly from the client.
• Development of a lightweight protocol for secure sharing of cell embeddings across research consortia without raw data exposure.

Main Results:

• CytoVerse successfully runs complex scFM models directly in the browser, eliminating the need for server-side processing.
• The framework efficiently searches extensive cell atlases using compressed indexing, demonstrating scalability.
• The developed protocol allows for privacy-preserving sharing of analytical results (embeddings) among collaborators.

Conclusions:

• CytoVerse offers a scalable and privacy-preserving solution for distributed single-cell analysis.
• The framework democratizes access to large-scale single-cell atlases by removing computational and privacy barriers.
• CytoVerse represents a significant advancement in enabling collaborative and accessible single-cell data exploration.