DISCERN: deep single-cell expression reconstruction for improved cell clustering and cell subtype and state detection

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Single-cell sequencing offers detailed cellular insights but is limited by sparse gene expression data.
• This data sparsity hinders accurate cell clustering and identification of cell types.
• Technical constraints in single-cell sequencing impact the depth of biological process analysis.

Purpose of the Study:

• To develop a novel deep generative network, DISCERN, for reconstructing missing single-cell gene expression.
• To enhance cell clustering and cell type identification using imputed gene expression data.
• To gain novel insights into cellular regulation and disease mechanisms through improved gene expression reconstruction.

Main Methods:

• Utilized a deep generative network (DISCERN) to infer missing gene expression from single-cell data.
• Employed a reference dataset for accurate gene expression reconstruction.
• Evaluated DISCERN's performance against competing algorithms for expression inference.

Main Results:

• DISCERN significantly improved cell clustering and cell type detection compared to existing methods.
• The method demonstrated robustness against batch effects while preserving biological variations.
• Identified novel COVID-19-associated T cell types (cytotoxic CD4+ and CD8+ Tc2 T helper cells).
• Achieved 80% AUROC in classifying COVID-19 severity using T cell fractions, indicating potential as a disease biomarker.

Conclusions:

• DISCERN is a flexible and effective tool for reconstructing missing single-cell gene expression.
• The algorithm can be readily integrated into existing single-cell sequencing workflows.
• DISCERN provides novel insights into disease mechanisms and cellular regulation.