gene-scoring 0.1.4

Gene set scoring package for single-cell RNA sequencing data with GPU acceleration.

Gene Scoring Package

This is a Python package implementing a novel gene set scoring methodology for single-cell RNA sequencing data analysis. It builds upon and extends techniques from widely used single-cell analysis toolkits such as Seurat and Scanpy, introducing several key improvements for more accurate gene set scoring.

Key Features

1. Weighted Gene Contributions

   • Each gene in the evaluated module is assigned an appropriate weight based on its correlation with the phenotype of interest
   • Weights can be normalized to sum to 1 (optional)

2. Individualized Background Selection

   • Background genes are selected independently for each gene of interest
   • Control features are selected from the same expression bin in the histogram
   • Background can be based on control sample, entire dataset, or a selected pool of genes

3. Variance-scaled Gene Contribution

   • Individual gene contributions are scaled by their variance under examined conditions
   • Helps account for gene expression variability in different conditions

4. Optimized Computation

   • GPU-accelerated score calculation for improved performance
   • Support for sparse matrices and chunked processing for large datasets
   • Efficient background gene selection algorithm

Installation

Basic Installation

pip install gene_scoring

Installation with GPU Support

pip install gene_scoring[gpu]

Mathematical Foundation

The scoring methodology is defined by the following formula:

S_c = Σ (w_l / (σ_c,l + ε)) * (X_c,l - X̄_l,control)

Where:

• n: number of genes in the gene list
• w_l: weight for gene l
• X_c,l: mean expression value of gene l in condition c
• X̄_l,control: average expression of control genes in the background
• σ_c,l: standard deviation of gene expression in condition c
• ε: small constant to avoid division by zero

Usage

from gene_scoring import GeneScorer

# Initialize the scorer
scorer = GeneScorer(
    normalize_weights=True,  # Normalize weights to sum to 1
    abs_diff=False,         # Use signed difference (not absolute)
    weighted=True           # Use weighted scoring
)

# Calculate scores
scores = scorer.calculate_scores(
    expression_matrix,    # Gene expression matrix (genes x cells)
    gene_list,           # List of genes to score
    weights=gene_weights  # Optional weights for each gene
)

Parameters

The main parameters for score calculation include:

• normalize_weights: If True, weights are normalized to sum to 1
• abs_diff: If True, use absolute difference between expression and background
• weighted: If True, use weighted scoring (otherwise all weights = 1)
• ctrl_size: Number of control genes to use per target gene
• gpu: Whether to use GPU acceleration
• chunk_size: Size of chunks for processing large datasets
• random_state: Random seed for reproducibility

Requirements

• Python ≥ 3.7
• NumPy ≥ 1.19.0
• Pandas ≥ 1.0.0
• SciPy ≥ 1.5.0
• CuPy ≥ 9.0.0 (optional, for GPU support)

License

This project is licensed under the MIT License - see the LICENSE file for details.