scvi-hub

scvi-hub enables efficient sharing and accessing of single-cell omics datasets using pre-trained probabilistic models, allowing immediate analysis of query datasets with reduced computational requirements.

Key Features:

• Pre-trained Model Distribution: Provides access to pre-trained probabilistic models for single-cell omics datasets without requiring data transfer, normalization, or integration procedures.
• Query Dataset Analysis: Supports visualization, imputation, annotation, outlier detection, and deconvolution of new query datasets using state of the art algorithms.
• Resource Efficiency: Reduces storage and compute requirements compared to standard approaches by leveraging pre-trained models rather than raw data processing.
• Reference Atlas Integration: Enables efficient analysis of existing references including large atlases such as the CZ CELLxGENE Discover Census.
• Extensible Repository: Allows inclusion of new datasets alongside already-loaded datasets within the scvi-tools open source environment and scverse integration.

Scientific Applications:

• Single-cell Data Visualization: Generates visualizations of query datasets using pre-trained model representations.
• Missing Data Imputation: Recovers missing gene expression values in single-cell datasets through probabilistic model inference.
• Cell Type Annotation: Assigns cell type labels to uncharacterized cells in query datasets using reference model knowledge.
• Anomaly Detection: Identifies outlier cells that deviate from expected patterns in single-cell data.
• Cellular Deconvolution: Estimates cell type proportions in mixed or bulk samples using single-cell reference models.
• Atlas-level Reuse: Facilitates new discoveries by enabling efficient reanalysis of existing large-scale reference collections.

Methodology:

scvi-hub is built within the scvi-tools open source environment and integrated into scverse, utilizing pre-trained probabilistic models to bypass data transfer, normalization, and integration steps typically required prior to analysis.