DURIAN: an integrative deconvolution and imputation method for robust signaling analysis of single-cell...

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) provides high-resolution gene expression data but often suffers from low read depth, leading to the loss of critical signaling gene information.
• This loss can hinder accurate analysis of cell-cell communication and biomarker discovery.

Purpose of the Study:

• To introduce DURIAN (Deconvolution and mUltitask-Regression-based ImputAtioN), an integrative computational method designed to recover lost gene expression data in scRNA-seq.
• To demonstrate the accuracy, robustness, and convergence of DURIAN through systematic benchmarking on synthetic and published datasets.

Main Methods:

• DURIAN employs a deconvolution and multitask-regression approach to impute missing gene expression values in scRNA-seq data.
• The method was validated using both simulated and real-world scRNA-seq datasets, comparing its performance against existing methods.

Main Results:

• DURIAN accurately recovers gene expression, outperforming existing methods in imputation tasks.
• The application of DURIAN significantly enhances single-cell clustering, improves low-dimensional embedding visualizations, and enables more accurate reconstruction of intercellular signaling networks.
• The study resolved inconsistencies observed when analyzing cell-cell communication using single-cell or bulk RNA sequencing data independently.

Conclusions:

• DURIAN is a robust and accurate computational tool for imputing gene expression in scRNA-seq data, addressing the limitations of low read depth.
• The method offers broad applicability for advancing biomarker discovery and cell signaling pathway analysis in single-cell transcriptomics research.
• DURIAN facilitates a more comprehensive understanding of cellular functions and interactions from scRNA-seq data.