EpiBench: Technical Requirements and Setup

This note outlines the technical requirements and setup process for EpiBench. Since this is an active project, requirements may evolve—check back for updates. Last updated: 2024-09-20

Hardware Requirements

EpiBench is designed to handle large genomic datasets and train sophisticated neural network models. As such, it has specific hardware recommendations for optimal performance:

Minimum Requirements

• CPU: 4+ cores (8+ threads)
• RAM: 16GB
• Storage: 50GB free space (SSD recommended)
• GPU: Not strictly required, but CPU-only training will be significantly slower

Recommended Specifications

• CPU: 8+ cores (16+ threads), modern AMD Ryzen or Intel i7/i9
• RAM: 32GB+
• Storage: 100GB+ SSD
• GPU: NVIDIA GPU with 8GB+ VRAM (RTX 2070 or better)
  • CUDA-compatible GPU required for GPU acceleration
  • RTX 3080/3090 or A100 recommended for large datasets or complex models

High-Performance Computing

For institutional users analyzing multiple datasets or training custom models, we recommend:

• Access to an HPC cluster with GPU nodes
• Job submission through SLURM, LSF, or similar scheduler
• Shared storage for large reference genomes and datasets

Software Dependencies

EpiBench relies on several key software components and libraries:

Core Requirements

• Python: Version 3.7+ (3.8 or 3.9 recommended)
• CUDA: Version 11.0+ (for GPU acceleration)
• cuDNN: Compatible with installed CUDA version

Key Python Libraries

Library	Purpose	Version
PyTorch	Deep learning framework	1.8.0+
Biopython	Genomic sequence processing	1.78+
pyBigWig	Histone modification data handling	0.3.18+
Pandas	Data manipulation	1.3.0+
NumPy	Numerical operations	1.20.0+
Scikit-learn	Traditional ML algorithms and metrics	0.24.0+
Matplotlib	Basic visualization	3.4.0+
Seaborn	Statistical visualization	0.11.0+
Plotly	Interactive visualization	5.1.0+

Environment Setup

We strongly recommend using a dedicated conda environment for EpiBench:

# Create a new conda environment
conda create -n epibench python=3.9
 
# Activate the environment
conda activate epibench
 
# Install PyTorch with CUDA support
conda install pytorch cudatoolkit=11.3 -c pytorch
 
# Install bioinformatics libraries
conda install -c bioconda biopython pybigwig pyfaidx
 
# Install data science tools
conda install pandas numpy scikit-learn matplotlib seaborn plotly
 
# Install additional dependencies
pip install captum  # For model interpretability

Data Requirements

EpiBench works with several data types, each with specific format requirements:

Reference Genome

• Complete reference genome in FASTA format
• Common choices include hg38 or hg19 for human data
• Must match the genome build used for your other data types

Epigenetic Data

• DNA Methylation: BED or CSV files with coordinates and methylation scores
• Histone Modifications: BigWig format signal tracks
  • Each histone mark should be in a separate file
  • Files should use the same genomic coordinate system

Sample Metadata

• CSV files with sample information
• Must include sample IDs that match filenames
• Additional clinical or experimental annotations as needed

Disk Space Considerations

Be aware that genomic analyses can require significant storage:

• Reference Genome: ~3GB (compressed)
• BigWig Files: ~1-5GB per histone mark per sample
• Methylation Data: ~0.5-2GB per sample
• Model Checkpoints: ~0.5-1GB per model
• Visualization Outputs: ~1-5GB depending on detail level

Installation Process

Option 1: Direct Installation

1. Clone the EpiBench repository:

   git clone https://github.com/username/epibench.git
   cd epibench