SIMBA: single-cell embedding along with features.

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Current single-cell analysis pipelines often rely on clustering and cell embeddings, limiting the explicit modeling of interactions between diverse feature types.
• Existing methods are task-specific, leading to fragmented approaches for distinct single-cell problems.

Purpose of the Study:

• To introduce SIMBA, a graph embedding method designed to overcome limitations in current single-cell analysis.
• To provide a unified framework for various single-cell analyses by jointly embedding cells and their defining features.

Main Methods:

• SIMBA employs a graph embedding approach to co-embed single cells with their defining features (genes, chromatin-accessible regions, DNA sequences) into a shared latent space.
• This method leverages the joint embedding to facilitate a range of downstream analyses.

Main Results:

• SIMBA enables the study of cellular heterogeneity and clustering-free marker discovery.
• The method supports gene regulation inference, batch effect removal, and multi-omics data integration within a single framework.
• Demonstrates simplification in developing new analyses and extending to novel single-cell modalities.

Conclusions:

• SIMBA offers a unified and flexible framework for diverse single-cell analysis tasks.
• The Python library implementation facilitates accessibility and further development in the field of single-cell genomics.