Debris Identification using Expectation Maximization (DIEM)

DIEM identifies and filters debris-contaminated droplets in droplet-based single-nucleus RNA-seq (snRNA-seq) datasets to reduce extranuclear RNA bias and improve downstream cell-type and gene-expression analyses.

Key Features:

• Contamination quantification and filtering: Uses a likelihood-based framework that models gene-expression distributions for debris and true cell types via Expectation Maximization to probabilistically identify and filter contaminated droplets.
• Improved data quality: Removes droplets with elevated extranuclear RNA contributions to produce higher-quality clusters and more reliable cell-type characterization, reducing artifactual clusters.
• Versatility across platforms: Applicable to droplet-based single-nucleus RNA-seq (snRNA-seq) and has been applied to single-cell RNA-seq (scRNA-seq) datasets.

Scientific Applications:

• Solid tissue profiling: Mitigates debris-driven signal in snRNA-seq studies of solid tissues to improve unbiased cell-type characterization.
• Gene-expression fidelity: Enhances the fidelity of gene-expression analyses relevant to tissue differentiation, cellular composition, and disease-associated transcriptional changes.

Methodology:

Applies a likelihood-based mixture model with Expectation Maximization to estimate gene-expression distributions for debris and genuine cellular populations, then classifies and filters droplets based on inferred contamination likelihood.