How pedrobarbosa/dress works
Project Type and Purpose
• Type: A Python-based, CLI-driven bioinformatics toolkit (open-source research software).
• Purpose: Generate and filter synthetic RNA exon-triplet datasets via genetic programming constrained by domain-specific grammars, guided by a deep-learning “oracle,” then provide evaluation utilities for motif analysis.
High-Level Components
• CLI Frontend (“dress” entry point)
– Subcommands: generate, filter
– Parses user arguments + YAML config files in dress/configs/*.yaml
• DatasetGeneration Module (dress/datasetgeneration)
– Preprocessing: gtf_cache + genome utilities
– Grammar loader (dress/datasetgeneration/grammars)
– Genetic programming controller (evolution.py, custom_callbacks, etc.)
– Underlying engine: GeneticEngine Python library
– Output: synthetic sequence dataset (FASTA/TSV)
• DatasetFiltering Module (dress/datasetfiltering)
– Input: synthetic or raw sequences
– Filters by splice-site probability, PSI, dPSI
– Validation, argument checking
– Output: filtered subset
• DatasetEvaluation Module (dress/datasetevaluation)
– Representation layer: APIs for sequences and motifs
– Motif DB loaders (ATtRACT, cisBP_RNA, oRNAment, RBNS_Encode)
– Enrichment and search utilities
– Output: evaluation reports, motif hit tables
• Data Stores & External Resources
– Genome reference (hg38 FASTA) + exon cache (GENCODE v44)
– Local motif databases under dress/datasetevaluation/representation/motifs/db
– Example inputs (data/examples)
• CI/Tests
– pytest suite under tests/
– GitHub Actions workflow (.github/workflows/run_tests.yml)
Relationships & Data Flow
→ Reads raw_input (data/examples or user-provided)
→ Loads YAML config
→ Calls DatasetGeneration
• Preprocesses via gtf_cache → Fetches exon sequences from genome + cache
• Loads grammar → Applies genetic operators → GeneticEngine evolves sequences
• At each generation, queries external oracle (SpliceAI/Pangolin) → returns scores
• Applies custom fitness, stopping, callbacks → produces synthetic dataset
→ Writes raw synthetic output (FASTA/TSV) to disk
b. User invokes “dress filter”
→ Reads synthetic or raw dataset
→ Applies probability/PSI thresholds → outputs filtered dataset
c. (Optional) User runs evaluation scripts in dress/datasetevaluation
→ Loads sequences + motif DBs
→ Runs enrichment/search → outputs visualizations or CSV summaries
Architectural Patterns & Design Principles
• Modular CLI → dispatch pattern (single entry point, subcommands)
• Clear separation of concerns: generation vs. filtering vs. evaluation
• Plugin-style grammars: grammars in their own package folder, easy to extend
• GeneticAlgorithm orchestration via GeneticEngine library
• YAML-based configuration → reproducibility
• Data-centric pipelines: each stage reads/writes files
• Test-driven development: extensive pytest suite
Elements
• Config files (dress/configs/*.yaml) feeding into both generate & filter
• GitHub Actions & pytest test suite as a side-component for CI portrayal