M3Drop: dropout-based feature selection for scRNASeq.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Gene expression analysis in genomics often requires focusing on a subset of relevant genes.
• Single-cell RNA sequencing (scRNASeq) generates large datasets where feature selection is crucial to remove technical noise.

Purpose of the Study:

• To introduce M3Drop, an R package for effective gene feature selection in scRNASeq data.
• To develop and evaluate novel methods for feature selection that account for zero-inflation (dropouts) in scRNASeq.

Main Methods:

• M3Drop implements existing feature selection techniques.
• Two new methods are introduced that leverage the high frequency of zero counts in scRNASeq data.
• Performance is assessed using simulated and real scRNASeq datasets.

Main Results:

• The novel methods within M3Drop demonstrate superior performance compared to existing approaches.
• M3Drop effectively identifies informative genes by addressing technical noise and dropouts.
• The R package is compatible with other widely used scRNASeq analysis tools.

Conclusions:

• M3Drop offers an improved approach to feature selection for scRNASeq data analysis.
• The package facilitates more accurate identification of biologically relevant genes from noisy datasets.
• M3Drop is readily available as an open-source R package on GitHub.