FastWilcox

Fast Wilcoxon rank-sum test for differential expression analysis in single-cell genomics.

This is a Python implementation of the wilcoxauc from the R' presto package, providing high-performance statistical testing for single-cell RNA-seq data.

Features

• Fast Wilcoxon rank-sum test with AUC calculation
• Multiple matrix formats supported: Dense, CSR, and CSC sparse matrices
• Optimized for single-cell data with efficient handling of sparse matrices
• FDR correction using Benjamini-Hochberg method
• Compatible with AnnData objects

Installation

cd python
pip install fastwilcox

For development:

pip install -e ".[dev]"

Usage

import anndata as ad
from fastwilcox import wilcoxauc

# Load your data
adata = ad.read_h5ad("your_data.h5ad")

# Run Wilcoxon test
results = wilcoxauc(
    x=adata.X,
    y=adata.obs['cell_type'].values,
    groups_use=None,  # Optional: test specific groups
    y_order=None,     # Optional: specify group order
    comparison=None   # Optional: pairwise comparison label
)

# Results is a list of WilcoxResult objects
for res in results:
    print(f"Feature {res.feature}, Group {res.group}: "
          f"logFC={res.log_fc:.3f}, AUC={res.auc:.3f}, "
          f"p-adj={res.padj:.3e}")

API Reference

wilcoxauc(x, y, groups_use=None, y_order=None, comparison=None)

Perform Wilcoxon rank-sum test for differential expression.

Parameters:

• x: Matrix (Dense array, CSR, or CSC sparse matrix) where rows are samples and columns are features
• y: List/array of group labels (strings)
• groups_use: Optional list of groups to test (default: all groups)
• y_order: Optional list specifying the order of groups
• comparison: Optional comparison pattern for pairwise tests

Returns:

• List of WilcoxResult objects with fields:
  • feature: Feature index
  • group: Group name
  • avg_expr: Average expression in group
  • log_fc: Log fold change (in-group mean - out-group mean)
  • ustat: U-statistic
  • auc: Area under ROC curve
  • pval: P-value
  • padj: FDR-adjusted p-value
  • pct_in: Percentage of cells expressing in group
  • pct_out: Percentage of cells expressing out of group
  • comparison: Comparison label, useful when only compare two group

Testing

Run tests:

cd python
pytest tests/

Run tests with coverage:

pytest tests/ --cov=fastwilcox --cov-report=html

Performance

This implementation is optimized for:

• Large sparse matrices (common in single-cell RNA-seq)
• Efficient group-wise operations
• Vectorized numpy operations

For the best performance with sparse data, use CSC format.

License

MIT License

Citation

If you use this software, please cite the fastwilcox project.