scCross: efficient search for rare subpopulations across multiple single-cell samples.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) data reveals cellular heterogeneity.
• Aggregating multiple scRNA-seq samples introduces batch effects, complicating analysis.
• Existing data integration methods can overcorrect, hindering rare cell identification.

Purpose of the Study:

• To present scCross, a novel biclustering method for identifying rare cell subpopulations across multiple scRNA-seq samples.
• To address the challenge of detecting rare cells in the presence of batch effects without prior data integration.

Main Methods:

• scCross employs a biclustering approach focusing on a global sum criterion for subpopulations.
• It identifies cells with specific marker genes by evaluating entire subpopulations, not pairwise cell comparisons.
• The method is designed to be robust against high variability and batch effects inherent in scRNA-seq data.

Main Results:

• scCross successfully identified rare cell subpopulations in lung cancer and human pancreas datasets without prior data integration.
• A cilium subpopulation with potential novel ciliary genes was detected in lung cancer cells, missed by alternative methods.
• The method demonstrated superior performance in identifying a target rare cell type in a controlled experiment with induced batch effects.

Conclusions:

• scCross effectively identifies rare cell subpopulations characterized by specific genes across multiple samples.
• The method is robust to batch effects, offering an advantage over traditional data integration techniques.
• scCross provides a valuable tool for exploring cellular heterogeneity in aggregated scRNA-seq studies.