Sfaira accelerates data and model reuse in single cell genomics.

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) data analysis is often performed independently, hindering knowledge integration from prior studies.
• Contextualizing new scRNA-seq datasets with public data requires extensive and time-consuming data preparation.

Purpose of the Study:

• To introduce sfaira, a comprehensive resource for scRNA-seq data and pre-trained models.
• To streamline the analysis and integration of public scRNA-seq datasets.
• To develop a robust cell type classification method adaptable to varying annotation granularities.

Main Methods:

• Developed sfaira, a data zoo for public scRNA-seq datasets and a model zoo with executable pre-trained models.
• Implemented an ontology-based system for metadata organization to facilitate data contributions.
• Adapted cross-entropy loss for cell type classification accommodating datasets with different levels of annotation detail.

Main Results:

• sfaira integrates a large collection of public scRNA-seq datasets.
• Pre-trained models within sfaira enable rapid analysis and contextualization.
• Demonstrated the effectiveness of sfaira by training models on 8 million cells across diverse anatomic partitions.

Conclusions:

• sfaira significantly reduces the effort required for scRNA-seq data analysis and integration.
• The platform facilitates reproducible research by providing accessible data and models.
• sfaira advances the field by enabling efficient utilization of large-scale scRNA-seq data.