CroweLM

Biotech AI Platform for Drug Discovery

CroweLM is an integrated AI platform combining NVIDIA BioNeMo NIMs with local LLMs for comprehensive drug discovery workflows. Built for researchers, scientists, and biotech professionals.

Features

• Drug Target Analysis - Automated UniProt, ChEMBL, and PubMed data integration
• Protein Structure Prediction - ESMFold via NVIDIA NIMs
• Molecule Generation - MolMIM for de novo drug design
• AI-Powered Research - Specialized biotech and research agents
• MLX Training - LoRA fine-tuning on Apple Silicon

Installation

# Install all components
pip install crowelm

# Or install specific packages
pip install crowelm-agents    # Biotech & research agents
pip install crowelm-nims      # NVIDIA NIMs integration
pip install crowelm-pipelines # Drug discovery workflows
pip install crowelm-training  # MLX fine-tuning (Apple Silicon)

Quick Start

Using the Biotech Agent

import asyncio
from crowelm.agents import BiotechAgent

async def main():
    async with BiotechAgent() as agent:
        # Analyze a drug target
        results = await agent.analyze_target("P15056")  # BRAF kinase
        print(results["analysis"])

asyncio.run(main())

Running the Drug Discovery Pipeline

import asyncio
from crowelm.pipelines import DrugDiscoveryPipeline

async def main():
    async with DrugDiscoveryPipeline() as pipeline:
        results = await pipeline.run_full_pipeline(
            target_id="P15056",
            generate_ligands=True,
            num_ligands=10
        )
        print(results["report"])

asyncio.run(main())

Using NVIDIA NIMs Directly

import asyncio
from crowelm.nims import NVIDIANIMs

async def main():
    async with NVIDIANIMs() as nims:
        # Predict protein structure
        structure = await nims.predict_structure_esmfold("MVLSPADKTNVKAAWGKVGAHAGEYGAEALERMFL")

        # Generate novel molecules
        molecules = await nims.generate_molecules(num_molecules=10, property_name="QED")

        # Ask scientific questions
        answer = await nims.science_chat("What is protein folding?")

asyncio.run(main())

CLI Usage

# Run biotech agent interactively
python -m crowelm.agents.biotech -i

# Analyze a drug target
python -m crowelm.agents.biotech --target P15056

# Run research agent
python -m crowelm.agents.research --topic "CRISPR drug discovery"

# Test NVIDIA NIMs
python -m crowelm.nims.nvidia --test

# Run drug discovery pipeline
python -m crowelm.pipelines.drug_discovery --target P15056

Docker

# Pull the agent runtime
docker pull crowelogic/crowelm-agents:latest

# Run interactively
docker run -it --rm \
  -e NVIDIA_API_KEY=$NVIDIA_API_KEY \
  crowelogic/crowelm-agents:latest

# Run a specific analysis
docker run --rm \
  -e NVIDIA_API_KEY=$NVIDIA_API_KEY \
  crowelogic/crowelm-agents:latest \
  python -m crowelm.agents.biotech --target P15056

npm CLI

# Install globally
npm install -g @crowe-logic/crowelm-cli

# Check system status
crowelm status

# Run agents
crowelm agent research --topic "drug discovery"
crowelm nvidia pipeline P15056

Environment Variables

Variable	Description	Required
NVIDIA_API_KEY	NVIDIA API key for NIMs	For NVIDIA features
CROWELM_MODEL_URL	Local model endpoint	No (default: localhost:12434)
CROWELM_MODEL_NAME	Model to use	No (default: crowelogic/crowelogic:v1.0)

Packages

Package	Description
crowelm	Meta-package installing all components
crowelm-core	Shared configuration and utilities
crowelm-agents	Biotech and research AI agents
crowelm-nims	NVIDIA BioNeMo NIMs integration
crowelm-pipelines	Drug discovery workflow pipelines
crowelm-training	MLX LoRA fine-tuning for Apple Silicon

Requirements

• Python 3.10+
• NVIDIA API key (for NIMs features)
• Docker (optional, for containerized deployment)
• Apple Silicon Mac (for MLX training)

License

MIT License - see LICENSE for details.

Author

Michael Crowe - Crowe Logic

---

CroweLM - Accelerating Drug Discovery with AI