rettxmutation 0.3.6

Extract Rett Syndrome mutations from genetic diagnosis report

RettX Mutation Analysis Library

A Python library for extracting and validating genetic mutations from clinical reports using an AI-powered agentic pipeline. Supports 6 Rett Syndrome-related genes across multiple languages, returning fully normalized HGVS nomenclature with genomic coordinates on both GRCh37 and GRCh38 assemblies.

🚀 Quick Start

Installation

pip install rettxmutation

Basic Usage

import asyncio
from rettxmutation import RettxServices, DefaultConfig

async def extract():
    config = DefaultConfig()  # loads from .env / environment variables

    with RettxServices(config) as services:
        result = await services.agent_extraction_service.extract_mutations(
            "The patient carries the mutation NM_004992.4:c.916C>T (p.Arg306Cys) in MECP2."
        )

        for key, mutation in result.mutations.items():
            pt = mutation.primary_transcript
            print(f"Gene:       {pt.gene_id}")
            print(f"Transcript: {pt.hgvs_transcript_variant}")
            print(f"Protein:    {pt.protein_consequence_tlr}")
            print(f"Type:       {mutation.variant_type}")
            for assembly, coord in mutation.genomic_coordinates.items():
                print(f"{assembly}:    {coord.hgvs}  (pos {coord.start:,}–{coord.end:,})")

asyncio.run(extract())

CLI Example

python examples/extract_from_file.py path/to/genetic_report.txt --verbose

✨ Key Features

• 🧬 Multi-Gene Support: MECP2, FOXG1, SLC6A1, CDKL5, EIF2B2, MEF2C — with curated RefSeq transcripts
• 🌍 Multilingual: Processes reports in English, Spanish, Greek, Turkish, and more
• 🤖 Agentic Extraction: Azure OpenAI-powered agent with tool-calling (gene registry lookup, variant validation, complex variant handling)
• ✅ HGVS Validation: Every mutation is validated via VariantValidator with automatic coordinate liftover (GRCh37 ↔ GRCh38)
• 🔒 PHI Redaction: Automatic removal of personal health information before LLM processing
• ⚡ Production Ready: Type-safe Pydantic v2 models, exponential backoff, connection pooling
• 🔄 Dual Assembly Output: Every mutation includes genomic coordinates on both GRCh37 and GRCh38
• 🏗️ Modular Architecture: Lazy-initialized services with dependency injection and context manager support

🧬 Supported Genes

Gene	Chromosome	Primary Transcript	Condition
MECP2	Xq28	NM_004992.4 (+NM_001110792.2)	Classic Rett Syndrome
FOXG1	14q12	NM_005249.5	Congenital variant Rett
SLC6A1	3p25.3	NM_003042.4	Myoclonic-atonic epilepsy
CDKL5	Xp22.13	NM_001323289.2	CDKL5 deficiency disorder
EIF2B2	14q24.3	NM_014239.4	Vanishing white matter disease
MEF2C	5q14.3	NM_002397.5 (+NM_001131005.2)	MEF2C haploinsufficiency

📊 Output Structure

The ExtractionResult contains:

ExtractionResult
├── mutations: Dict[str, GeneMutation]    ← keyed by GRCh38 genomic HGVS
├── genes_detected: List[str]             ← e.g. ["MECP2"]
├── extraction_log: List[str]             ← agent reasoning trace
└── tool_calls_count: int                 ← total tool invocations

Each GeneMutation provides:

GeneMutation
├── genomic_coordinates:
│   ├── GRCh38: { assembly, hgvs, start, end, size }
│   └── GRCh37: { assembly, hgvs, start, end, size }
├── variant_type: "SNV" | "deletion" | "duplication" | "insertion" | "indel"
├── primary_transcript:
│   ├── gene_id, transcript_id
│   ├── hgvs_transcript_variant      ← e.g. NM_004992.4:c.916C>T
│   ├── protein_consequence_tlr      ← e.g. NP_004983.1:p.(Arg306Cys)
│   └── protein_consequence_slr      ← e.g. NP_004983.1:p.(R306C)
└── secondary_transcript (optional)

🛠️ Requirements

Python Version

• Python 3.8 or higher

Azure Services

Service	Required?	Purpose
Azure OpenAI	✅ Required	Agentic mutation extraction
Azure AI Search	Optional	Semantic search for keyword detection
Azure Cognitive Services	Optional	Text analytics enrichment

Environment Variables

# Required — Azure OpenAI
RETTX_OPENAI_ENDPOINT=https://your-openai.openai.azure.com/
RETTX_OPENAI_KEY=your-openai-key
RETTX_OPENAI_MODEL_NAME=gpt-4o           # deployment name

# Optional — Agent model (defaults to RETTX_OPENAI_MODEL_NAME if not set)
RETTX_OPENAI_AGENT_DEPLOYMENT=gpt-4o     # agent-specific deployment
RETTX_OPENAI_AGENT_MODEL_VERSION=2024-11-20

# Optional — Embeddings
RETTX_EMBEDDING_DEPLOYMENT=text-embedding-ada-002

# Optional — Azure AI Search
RETTX_AI_SEARCH_SERVICE=your-search-service
RETTX_AI_SEARCH_API_KEY=your-search-key
RETTX_AI_SEARCH_INDEX_NAME=your-index-name

# Optional — Azure Cognitive Services
RETTX_COGNITIVE_SERVICES_ENDPOINT=https://your-cognitive-services.cognitiveservices.azure.com/
RETTX_COGNITIVE_SERVICES_KEY=your-cognitive-services-key

📋 Processing Pipeline

The agentic extraction pipeline works as follows:

Input Text (any language)
    │
    ▼
┌──────────────────────┐
│  PHI Redaction        │  Remove patient names, DOBs, IDs
└──────────┬───────────┘
           ▼
┌──────────────────────┐
│  AI Agent (OpenAI)    │  Reads text, identifies mutations
│                      │
│  Tools available:    │
│  • lookup_gene_registry  → gene info + RefSeq transcripts
│  • validate_variant      → HGVS validation + coordinates
│  • validate_complex      → CNV / genomic coordinate validation
└──────────┬───────────┘
           ▼
┌──────────────────────┐
│  Structured Output    │  ExtractionResult with validated
│                      │  GeneMutation objects
└──────────────────────┘

Key capabilities:

• Ensembl → RefSeq remapping: Handles reports using Ensembl transcripts (ENST*) by using genomic coordinates
• Minus-strand awareness: Correctly complements alleles for genes on the reverse strand
• Old nomenclature: Normalizes legacy formats (e.g., 502C->T, R168X) to current HGVS

💻 Available Services

All services are lazily initialized via RettxServices:

with RettxServices(config) as services:
    # Core extraction
    services.agent_extraction_service    # AI-powered mutation extraction
    
    # Validation & analysis
    services.variant_validator_service   # HGVS validation + coordinate liftover
    services.mutation_tokenizator        # Mutation string tokenization
    
    # Search & embeddings
    services.embedding_service           # Azure OpenAI embeddings
    services.ai_search_service           # Azure AI Search integration
    services.keyword_detector_service    # Multi-layer keyword detection

Direct Variant Validation

with RettxServices(config) as services:
    vvs = services.variant_validator_service

    # Validate a transcript-level variant
    result = vvs.get_gene_mutation_from_transcript("NM_004992.4:c.916C>T")
    print(result.primary_transcript.protein_consequence_tlr)
    # → NP_004983.1:p.(Arg306Cys)

    # Validate a complex / genomic variant
    result = vvs.create_gene_mutation_from_complex_variant(
        assembly_build="GRCh38",
        assembly_refseq="NC_000023.11",
        variant_description="NC_000023.11:g.154030912G>A",
        gene_symbol="MECP2"
    )

Custom Configuration

class MyConfig:
    """Custom configuration for production (e.g., from Key Vault)."""
    RETTX_OPENAI_ENDPOINT = "https://my-openai.openai.azure.com/"
    RETTX_OPENAI_KEY = get_secret("openai-key")
    RETTX_OPENAI_MODEL_NAME = "gpt-4o"
    # Only set fields needed for the services you use

with RettxServices(MyConfig()) as services:
    result = await services.agent_extraction_service.extract_mutations(text)

🧪 Golden Test Suite

The library includes a comprehensive golden test suite with 11 real-world genetic reports:

Gene	Variant Type	Language	Key Feature
MECP2	SNV, splicing, deletion, duplication	EN, ES, EL, TR	Multiple transcripts
FOXG1	Frameshift deletion	TR	Non-MECP2 gene
SLC6A1	Whole-gene CNV (~20kb)	ES	Copy number variant

Run golden tests:

# Mock mode (no API calls, uses recorded responses)
python -m pytest tests/golden/ --golden-mode=mock -v

# Live mode (calls real APIs)
python -m pytest tests/golden/ --golden-mode=live -v

🎯 Use Cases

• 🏥 Clinical Genetics: Extract mutations from diagnostic reports in any language
• 🔬 Research: Analyze genetic data across Rett Syndrome and related conditions
• 📊 Patient Registries: Populate genetic databases with normalized HGVS nomenclature
• 🤖 Bioinformatics Pipelines: Integrate as a library or via the CLI example
• 📱 Clinical Applications: Build tools with structured mutation data (dual-assembly coordinates)

🔧 Reliability

• Exponential Backoff: Automatic retry for VariantValidator and OpenAI API calls
• Graceful Degradation: Optional services (AI Search, Cognitive Services) degrade gracefully
• PHI Redaction: Patient data is stripped before any LLM processing
• Type Safety: Pydantic v2 models with runtime validation
• Context Manager: Automatic resource cleanup via with statement
• Comprehensive Logging: Structured extraction logs with tool call traces

🤝 Contributing

We welcome contributions! Please see our GitHub repository for:

• Issue reporting
• Feature requests
• Pull request guidelines
• Development setup instructions

📄 License

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

🆘 Support

• Issues: GitHub Issues
• Documentation: API Documentation
• Contact: [e-mail]

🔮 Roadmap

• Additional Genes: Expand the gene registry beyond the current 6 genes
• Batch Processing: Process multiple reports in parallel with rate limiting
• Confidence Scoring: Per-mutation confidence metrics based on report quality
• Structured Report Parsing: Native support for VCF, JSON, and HL7 FHIR formats
• Cloud Deployment: Docker containers and Azure deployment templates