Boolean Correlation Networks Analysis Tool

Overview

This project implements a computational method for analyzing Boolean correlation networks of gene expression and methylation data. It combines and optimizes two existing frameworks - StepMiner and BooleanNet - with CUDA acceleration to efficiently process large-scale biological datasets.

Key Features

• Discretization of gene expression data using an optimized StepMiner implementation
• CUDA-accelerated BooleanNet implementation for Boolean implication network analysis
• Data compression techniques for efficient memory usage
• Random model generation for statistical validation
• Support for large-scale genomic datasets

Technical Implementation

StepMiner Optimization

• 2-bit compression for discretized values
• Bitwise operations for quadrant population computation
• Hardware-accelerated bit counting using intrinsics

CUDA Implementation

• Stream processing architecture for parallel computation
• Device-wide synchronized memory access
• Optimized matrix operations using triangular symmetry
• Custom memory management for large datasets

Statistical Validation

• Random model generation using C++11 mt19937
• Row and column permutation strategies
• Significance testing against null models

Dependencies

• CUDA Toolkit
• C++11 compatible compiler
• STL libraries

Usage

1. Prepare your expression/methylation data in matrix format

2. Configure analysis parameters:

   • BNstatThresh: Statistical threshold for implications
   • BNpvalThresh: P-value threshold
   • SMGap: StepMiner gap parameter (default: 0.5)

3. Run the analysis pipeline:

# Example command
./analyze_network --input data.csv --output results.txt --threads 64

Validation Results

The tool has been validated using the Cancer Genome Atlas Colon Adenocarcinoma dataset:

• Original implications: 421,206
• After permutation testing: 106,632

Performance

The CUDA implementation provides significant performance improvements over CPU-based processing, particularly for large datasets. Performance scales with the number of CUDA cores available.

Citation

If you use this tool in your research, please cite:

Greco, L.V. (2023). Implementation of random models on Boolean correlation networks of gene expression and methylation. University of Catania.

Contributing

Contributions are welcome! Please read the contributing guidelines before submitting pull requests.

Acknowledgments

• Giovanni Micale (Advisor)
• Alfredo Ferro (Co-advisor)
• Salvatore Alaimo
• Ilaria Cosentini
• University of Catania, Department of Mathematics and Computer Science