scTrimClust: a fast approach to robust scRNA-seq analysis using trimmed cell clusters.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell sequencing analysis relies heavily on unsupervised clustering.
• Outlier cells in clustering can skew downstream analyses like marker gene detection.
• Identifying and managing these non-representative cells is crucial for accurate interpretation.

Purpose of the Study:

• To introduce scTrimClust, a novel and efficient method for identifying extreme cells in single-cell data.
• To assess the impact of non-representative cells on scRNA-seq analysis outcomes.
• To provide a tool for evaluating the influence of different analysis parameters.

Main Methods:

• scTrimClust measures cell distances to nearest neighbors in high-dimensional gene expression space.
• Cells with minimum neighbor distances above a cluster-specific quantile threshold are flagged as extreme.
• The method was evaluated using two example datasets.

Main Results:

• scTrimClust effectively identifies cells with non-representative expression profiles.
• The study demonstrates how these extreme cells can influence analysis results.
• The approach aids in comparing the effects of various scRNA-seq analysis parameters.

Conclusions:

• scTrimClust offers a fast and effective way to refine single-cell datasets.
• Removing extreme cells can lead to more robust and reliable downstream analyses.
• The scTrimClust method is implemented in the R-package RepeatedHighDim.