Skip to content

Agentic-SpliceAI Setup Guide

Quick setup instructions for development and deployment

Overview

Agentic-SpliceAI builds upon the Meta-SpliceAI framework, combining:

  1. Extensible Base Layer - Foundation models (SpliceAI, OpenSpliceAI, + extensible)
  2. Adaptive Meta-Learning - Foundation-Adaptor framework via multimodal deep learning
  3. Agentic Workflows - AI agents for validation and evidence synthesis

Environment Setup

1. Create Dedicated Conda Environment

cd /Users/pleiadian53/work/agentic-spliceai

# Create environment from yml
mamba env create -f environment.yml

# Activate environment
mamba activate agentic-spliceai

2. Install Package in Development Mode

# Install agentic-spliceai package
pip install -e .

# Verify installation
python -c "import agentic_spliceai; print('✓ Package installed')"

3. Test CLI Commands

# Test CLI entry points
agentic-spliceai --help
agentic-spliceai-server --help

Architecture

Three-Layer System

┌─────────────────────────────────────────────────────────────┐
│                    AGENTIC-SPLICEAI                         │
├─────────────────────────────────────────────────────────────┤
│                                                             │
│  ┌──────────────────┐  ┌──────────────────┐  ┌──────────┐ │
│  │ Nexus Research   │  │ Meta-SpliceAI    │  │ Agentic  │ │
│  │ Agent            │  │ Prediction       │  │ Workflow │ │
│  ├──────────────────┤  ├──────────────────┤  ├──────────┤ │
│  │ • Literature     │  │ • Base SpliceAI  │  │ • LLM    │ │
│  │   synthesis      │  │ • Meta-learning  │  │   agents │ │
│  │ • arXiv/PubMed   │  │ • Ensemble       │  │ • Tool   │ │
│  │ • LaTeX reports  │  │   models         │  │   use    │ │
│  │ • Web search     │  │ • Feature eng.   │  │ • Multi- │ │
│  │                  │  │                  │  │   agent  │ │
│  └──────────────────┘  └──────────────────┘  └──────────┘ │
│                                                             │
└─────────────────────────────────────────────────────────────┘

Dependency Breakdown

From Nexus (Research): - LLM clients: openai, anthropic, mistralai, aisuite - Research tools: tavily-python, wikipedia, beautifulsoup4 - PDF generation: weasyprint, pandoc, tectonic, pypandoc - Web framework: fastapi, uvicorn, pydantic

From Meta-SpliceAI (Prediction): - Deep learning: tensorflow, pytorch, keras - SpliceAI: spliceai==1.3.1 - Genomics: biopython, pysam, pybedtools, pybigwig - Genomic formats: bcbio-gff, gffutils, gtftools, pyfastx - ML optimization: optuna, hyperopt, hpbandster - Feature engineering: category-encoders, feature-engine

Shared (Both): - Data: pandas, polars, numpy, duckdb - Visualization: matplotlib, seaborn, plotly - ML: scikit-learn, lightgbm, xgboost - Utilities: tqdm, joblib, requests

Development Workflow

Phase 1: Port Nexus Research Capability

# Copy Nexus research agent code
cp -r /Users/pleiadian53/work/agentic-ai-lab/src/nexus/agents/research/* \
      /Users/pleiadian53/work/agentic-spliceai/agentic_spliceai/research/

# Test research capability
python -c "from agentic_spliceai.research import pipeline; print('✓ Research imported')"

Phase 2: Integrate Meta-SpliceAI Prediction

# Copy meta-spliceai core modules
cp -r /Users/pleiadian53/work/meta-spliceai/meta_spliceai/splice_engine/* \
      /Users/pleiadian53/work/agentic-spliceai/agentic_spliceai/prediction/

# Test prediction capability
python -c "from agentic_spliceai.prediction import base_model; print('✓ Prediction imported')"

Phase 3: Build Agentic Workflow

# Combine research + prediction with LLM agents
# Create workflow that:
# 1. Uses Nexus to research latest splice site methods
# 2. Uses Meta-SpliceAI to make predictions
# 3. Uses LLM to interpret results and generate insights

Environment Comparison

Feature agentic-ai agentic-spliceai
Purpose General research Splice site research
Nexus ✓ Full ✓ Ported
Deep Learning ✗ ✓ TensorFlow/PyTorch
Genomics ✗ ✓ Full bioinformatics
SpliceAI ✗ ✓ Base + Meta
Size ~2GB ~5GB (with models)

Next Steps

  1. ✅ Create environment: mamba env create -f environment.yml
  2. ✅ Install package: pip install -e .
  3. ⏳ Port Nexus research agent
  4. ⏳ Integrate Meta-SpliceAI prediction
  5. ⏳ Build unified agentic workflow
  6. ⏳ Test end-to-end pipeline
  7. ⏳ Create example notebooks
  8. ⏳ Write documentation
  9. ⏳ First commit to GitHub

Troubleshooting

TensorFlow/PyTorch Conflicts

If you encounter GPU/CUDA issues: 

# CPU-only versions (lighter)
pip install tensorflow-cpu
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu

Genomics Tools

Some genomics tools require system dependencies: 

# macOS
brew install htslib bedtools

# Ubuntu/Debian
sudo apt-get install libhts-dev bedtools

Memory Issues

Large models may require significant RAM: - Minimum: 16GB RAM - Recommended: 32GB RAM - With GPU: 64GB+ RAM for large-scale training

Verify Installation

Run the setup verification script:

python scripts/setup/verify_setup.py

This checks: - Required packages installed - Import paths working - Data directories accessible - Environment variables set

Resources

  • Documentation:
  • README.md - Project overview and vision
  • QUICKSTART.md - Quick start guide
  • docs/STRUCTURE.md - Complete project structure
  • docs/ - Comprehensive documentation
  • Meta-SpliceAI: https://github.com/pleiadian53/meta-spliceai
  • Agentic AI Lab: https://github.com/pleiadian53/agentic-ai-lab