gtars Module Vision

Conceptual Organization

The gtars modules follow a clear separation of concerns:

1. gtars-overlaprs

Core overlap computation infrastructure

• All interval overlap algorithms live here
• Single source of truth for overlap operations
• Other modules should never reimplement overlap logic

2. gtars-scoring

Wraps overlaprs to produce X-by-peak matrices

• Handles all matrix generation use cases:
  • Cell-by-peak (single-cell)
  • Sample-by-peak (bulk)
  • Pseudobulk-by-peak (aggregated)
• Focuses on matrix data structures and I/O formats

3. gtars-tokenizers

Wraps overlaprs to produce tokens for ML

• Converts genomic regions to vocabulary tokens
• Designed for transformer models and ML pipelines
• Manages token vocabularies and encoding strategies

Key Principle

All overlap computation flows through gtars-overlaprs. Higher-level modules add domain-specific value:

• scoring adds matrix generation
• tokenizers adds ML tokenization
• Both delegate overlap computation to overlaprs