Model-based deep embedding for constrained clustering analysis of single cell RNA-seq data.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Clustering is essential for single-cell studies, but unsupervised methods struggle with high-dimensional scRNA-Seq data and dropout events.
• Existing unsupervised clustering approaches often yield biologically uninterpretable clusters, complicating cell type assignment and requiring laborious manual parameter tuning.

Purpose of the Study:

• To introduce scDCC, a novel clustering method that integrates domain knowledge into the clustering process for scRNA-Seq data.
• To enhance the biological interpretability of clustering results and simplify downstream analyses like cell type assignment.

Main Methods:

• Developed scDCC, a principled clustering approach that incorporates prior biological knowledge.
• Applied scDCC to diverse scRNA-Seq datasets encompassing thousands to tens of thousands of cells.

Main Results:

• scDCC significantly improves clustering performance on scRNA-Seq data.
• The method facilitates more interpretable clusters compared to purely unsupervised methods.
• Enhanced cluster interpretability aids in accurate cell type assignment and downstream analysis.

Conclusions:

• Integrating domain knowledge into clustering offers a principled solution for scRNA-Seq data challenges.
• scDCC provides a robust and efficient tool for improving the quality and interpretability of single-cell data clustering.
• This approach streamlines the analysis of scRNA-Seq data, leading to more reliable biological insights.