jab57/mcp-server-project

GREmLN MCP Server

Gene Regulatory network-informed Large Language Model (GREmLN) MCP Server for analyzing regulatory networks and gene relationships.

Setup Instructions

1. Environment Setup

# Create and activate virtual environment
python -m venv gremln-env
source gremln-env/Scripts/activate  # Windows
# source gremln-env/bin/activate    # Linux/Mac

# Install dependencies
pip install -r requirements.txt

2. Install GREmLN Dependencies

⚠️ scGraphLLM Installation Required

The scGraphLLM package is not available on PyPI. You need to:

1. Find the source repository - Check the GREmLN paper or contact authors
2. Install from source:

   # If available on GitHub:
   pip install git+https://github.com/[author]/scGraphLLM.git
   
   # Or clone and install:
   git clone https://github.com/[author]/scGraphLLM.git
   cd scGraphLLM
   pip install -e .
   

3. Download Model and Data

You'll need:

• GDTransformer model checkpoint: models/model.ckpt
• Regulatory networks: Cell-type specific .tsv files in models/networks/

Data Sources: The regulatory networks are derived from CellxGene single-cell RNA-seq data processed through a comprehensive ARACNe-based pipeline. See gremln_source/scripts/README.md for complete preprocessing details and GREmLN_Analysis_Pipeline.md for data lineage documentation.

Expected structure:

GREmLN/
├── models/
│   ├── model.ckpt
│   └── networks/
│       ├── cd14_monocytes/
│       │   ├── network.tsv          # Original ARACNe output
│       │   └── network_index.pkl    # Optimized cache (generated)
│       ├── cd16_monocytes/
│       ├── cd4_t_cells/
│       ├── cd8_t_cells/
│       ├── epithelial_cell/
│       └── ... (15 total cell types)

Data Processing Pipeline:

CellxGene Data → QC/Metacells → ARACNe Networks → TSV Files → Pickle Caches

4. Test Installation

python gremln_mcp_server.py

Features

Current Capabilities

• Network Analysis: Analyze gene positions in regulatory networks
• Cell-Type Specific: Support for 15 different cell types
• Gene Similarity: Find genes with similar regulatory patterns
• Network Statistics: Get detailed network information

Cell Types Supported

15 Major Human Cell Types (derived from CellxGene datasets):

Immune & Blood Cell Types:

• CD14 Monocytes - Classical circulating monocytes
• CD16 Monocytes - Non-classical patrolling monocytes
• CD20 B Cells - B lymphocytes
• CD4 T Cells - Helper T cells
• CD8 T Cells - Cytotoxic T cells
• Erythrocytes - Red blood cells
• NK Cells - Natural killer cells
• NKT Cells - Natural killer T cells
• Monocyte-derived Dendritic Cells - Antigen-presenting cells

Tissue & Organ Cell Types:

• Epithelial Cells - Barrier tissue cells
• Hepatocytes - Liver cells (drug metabolism)
• Cardiomyocytes - Heart muscle cells
• Neurons - Brain and nervous system cells
• Fibroblasts - Connective tissue cells
• Endothelial Cells - Blood vessel lining cells

Each cell type contains ARACNe-generated gene regulatory networks with 403-183,247 regulatory edges based on the top 1024 highly variable genes.

MCP Tools Available

1. analyze_gene_network

   • Analyze gene's regulatory context
   • Parameters: gene, cell_type

2. find_similar_genes

   • Find genes with similar regulatory patterns
   • Parameters: gene, cell_type, top_n

3. get_network_info

   • Get network statistics
   • Parameters: cell_type

Architecture

• gremln_mcp_server.py: Main MCP server
• agents/: Multi-agent framework (planned)
• utils/: Utility functions
• models/: Model checkpoints and networks
• complete_gene_service.py: Gene annotation service

Status

🟡 Partial Implementation

• ✅ Basic MCP server structure
• ✅ Core dependencies installed
• ⚠️ Requires scGraphLLM installation
• ⚠️ Requires model/network data
• 🔄 Multi-agent framework pending

Next Steps

1. Locate and install scGraphLLM package
2. Download GREmLN model checkpoint
3. Download regulatory network data
4. Implement advanced multi-agent features
5. Add visualization capabilities