A downsampling method enables robust clustering and integration of single-cell transcriptome data.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) data often contains noise, sampling biases, and batch effects.
• These technical variations can obscure true biological signals, hindering downstream analyses.
• Accurate noise and bias adjustment is crucial for reliable biological discoveries.

Purpose of the Study:

• To develop a model-based downsampling algorithm to address noise and biases in scRNA-seq data.
• To create a method that preserves biological covariance while reducing technical errors.
• To enhance the accuracy of cell clustering, gene selection, and data integration.

Main Methods:

• Proposed a novel algorithm named minimal unbiased representative points (MURP^XMBD).
• MURP^XMBD employs a model-based downsampling approach.
• The algorithm focuses on reducing gene-wise random errors while maintaining biological covariance structure.

Main Results:

• MURP^XMBD successfully retrieves representative points from scRNA-seq data.
• The algorithm provides an unbiased representation of cell populations.
• Validation on benchmark datasets demonstrated improved clustering quality and accuracy.
• Enhanced performance in dataset integration algorithms was also observed.

Conclusions:

• MURP^XMBD is an effective noise-reduction method for scRNA-seq analysis.
• The algorithm facilitates the discovery of novel cell types.
• MURP^XMBD supports more robust downstream analyses in biomedical studies.