DISC: a highly scalable and accurate inference of gene expression and structure for single-cell transcriptomes using...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) is crucial for understanding cellular heterogeneity.
• Gene expression data in scRNA-seq is often sparse due to technical dropouts.
• These dropouts can lead to inaccurate gene expression profiles and misclassification of cell types.

Purpose of the Study:

• To develop a robust method for inferring gene expression obscured by dropouts in scRNA-seq data.
• To improve the accuracy of gene expression profiles and cell type identification.
• To provide a scalable and reliable tool for analyzing sparse scRNA-seq datasets.

Main Methods:

• Development of DISC, a novel deep learning network utilizing semi-supervised learning.
• Training and validation of DISC on ten real-world scRNA-seq datasets.
• Comparative analysis against seven state-of-the-art imputation methods.

Main Results:

• DISC consistently outperformed seven existing imputation approaches across all tested datasets.
• The method demonstrated significant improvements in recovering gene expression patterns.
• Enhanced gene and cell structure identification was observed, leading to better cell type classification.

Conclusions:

• DISC offers a powerful and reliable solution for addressing dropout issues in scRNA-seq data.
• The method's performance, scalability, and applicability make it a valuable tool for single-cell data analysis.
• DISC enhances the accuracy of gene expression recovery and cell type identification, advancing the field of single-cell genomics.