Enhancing droplet-based single-nucleus RNA-seq resolution using the semi-supervised machine learning classifier DIEM.

Area of Science:

• Genomics
• Bioinformatics
• Molecular Biology

Background:

• Single-nucleus RNA sequencing (snRNA-seq) enables gene expression analysis in individual nuclei for cell type characterization in solid tissues.
• Ambient RNA contamination is a common issue in snRNA-seq, potentially causing biased results and spurious cell type identification.

Purpose of the Study:

• To introduce Debris Identification using Expectation Maximization (DIEM), a new computational method for quantifying and filtering ambient RNA contamination in snRNA-seq data.
• To demonstrate DIEM's effectiveness in improving the quality of downstream analyses, including cell type identification.

Main Methods:

• DIEM employs a likelihood-based approach using Expectation-Maximization (EM) to model gene expression distributions of debris and cell types.
• The method was evaluated on three diverse snRNA-seq datasets: human preadipocytes, mouse brain, and human adipose tissue.
• DIEM's performance was compared against existing state-of-the-art filtering techniques.

Main Results:

• All tested datasets exhibited extranuclear RNA contamination, which existing methods failed to adequately address, leading to spurious cell types.
• DIEM demonstrated superior removal of droplets with high levels of extranuclear RNA compared to current methods.
• Filtering with DIEM resulted in higher quality cell clusters and more accurate downstream analyses.

Conclusions:

• DIEM provides a fast and effective solution for removing debris-contaminated droplets in single-cell and single-nucleus RNA sequencing data.
• The method enhances the reliability and accuracy of downstream analyses by producing cleaner datasets.
• The DIEM algorithm and its code are publicly available for researchers.