scKWARN

scKWARN performs robust normalization of single-cell RNA-seq data by generating kernel-weighted pseudo expression profiles across fuzzy technical neighbors to correct sequencing-depth, capture-efficiency, and dropout-related technical biases while preserving biological heterogeneity.

Key Features:

• Kernel-weighted averaging: Uses a kernel smoother to borrow information across similar cells to generate pseudo expression profiles.
• Fuzzy technical neighbors: Identifies fuzzy technical neighbors for each cell to capture local technical patterns without strict neighborhood boundaries.
• Pseudo expression profiles: Constructs per-cell pseudo profiles by aggregating neighbor information to mitigate noise and bias.
• Bimodality-based reference alignment: Compares pseudo profiles to reference profiles derived from cells exhibiting identical bimodality patterns.
• Normalization factor estimation: Determines per-cell normalization factors from profile alignment to correct technical confounders.
• Distribution-free assumptions: Operates without relying on restrictive assumptions about data distribution or count-depth relationships.
• Preservation of biological heterogeneity: Removes technical biases while retaining genuine biological variability.
• Empirical validation: Evaluated on simulated and real datasets with demonstrated improvements in bias correction and retention of biological variability.
• Downstream compatibility: Produces normalized data suitable for clustering, differential expression analysis, and trajectory inference.

Scientific Applications:

• Technical bias correction: Corrects for sequencing depth, capture efficiency, and dropout effects in single-cell RNA-seq datasets.
• Cell clustering: Preserves heterogeneity to improve identification of cell types and states.
• Differential expression analysis: Provides normalized expression profiles for robust differential testing.
• Trajectory and lineage inference: Maintains biological signals required for trajectory reconstruction.
• Bimodal expression handling: Aligns cells by bimodality patterns to accommodate subpopulation-specific expression modes.

Methodology:

Per-cell pseudo expression profiles are generated by kernel smoothing over fuzzy technical neighbors; pseudo profiles are compared to reference profiles from cells with identical bimodality patterns; per-cell normalization factors are determined from these comparisons.