scverse 2024 recap: Joint TileDB-CZI Workshop

Recently in Munich, Germany, the inaugural scverse conference brought together the single cell community: researchers and toolkit developers for presentations, workshops, and poster sessions. scverse is a consortium of open source developers who maintain a suite of popular python packages for single-cell analysis, including anndata, scanpy, and more. The goal of this conference was to foster collaboration and standardization across the single cell community. TileDB was proud to sponsor and participate in the event.

The Chan Zuckerberg Initiative (CZI) and TileDB led a workshop demonstrating the impact of TileDB for single-cell research, specifically in the functionality of CZI’s CELLxGENE project. TileDB, the foundational database designed for scientific discovery, developed SOMA functionality alongside CZI to enable scalable queries of single-cell datasets traditionally inaccessible due to resource constraints and language-based silos. TileDB’s single-cell capabilities have since enabled the CZI CELLxGENE Census to house millions of cells and annotations that are accessible to the broader scientific community.

The joint workshop at scverse showcased the ease of training ML models on large-scale single-cell data. The session included a walkthrough on training a PyTorch model on CZI’s atlas-scale CELLxGENE census by leveraging the TileDB engine, as well as cell similarity searches on this same census using experimental single-cell data stored in TileDB. We highlighted TileDB's integration with scverse packages through an I/O API, facilitating effortless conversion of query results into scverse formats. This interoperability underscores TileDB's dedication to building software that harmonizes with the broader single-cell analysis ecosystem.

Conversations at the conference ranged widely, with machine learning and spatial transcriptomics leading the charge. The community provided a fantastic opportunity to share our latest developments and connect with users, learning firsthand how they’re utilizing TileDB’s single-cell functionality, which addresses a key need for performant out-of-core access to single-cell data. This significance is maximized when users can integrate with the scverse libraries they love, enabling advanced single-cell analysis workflows.

We are thankful to both the scverse core team for organizing the conference, as well as the single cell community for their engagement and ongoing contributions to the field.