PKPD Platform

A Python-first platform for pharmacokinetics, pharmacodynamics, and toxicity prediction using SciPy alternatives.

🎯 Three Main Goals

1. Predict human PK (CL, F, t½) and TI (QTc, bone marrow tox, DILI)
2. Build cross-species PK/PD virtual models (rodent/dog/NHP) to estimate NOAEL at 14d/1m/9m
3. Guide compound design toward human PK/PD/toxicity targets

🚀 Key Features

• ✅ Human PK/TI Prediction - Using RDKit + ADMET-AI + custom models
• ✅ Cross-species PK/PD Models - Using SciPy ODE solvers + allometric scaling
• ✅ Compound Design & Optimization - Using SciPy optimization + genetic algorithms
• ✅ Production Ready - No complex compilation, GKE deployment ready

📦 Installation

Local Development

# Clone repository
git clone <repository-url>
cd PKPD

# Install dependencies
pip install -e .

Docker Deployment

# Build image
docker build -t pkpd:latest .

# Run container
docker run -p 8000:8000 pkpd:latest

GKE Deployment

# Build and push to GCR
docker build -t gcr.io/PROJECT_ID/pkpd:latest .
docker push gcr.io/PROJECT_ID/pkpd:latest

# Deploy to GKE
kubectl apply -f k8s/

🧪 Quick Start

Run the Demonstration Notebook

jupyter notebook notebooks/01_getting_started.ipynb

This notebook demonstrates all three goals:

• Goal 1: Human PK/TI Prediction
• Goal 2: Cross-species PK/PD Models & NOAEL Estimation
• Goal 3: Compound Design & Optimization

CLI Usage

# Predict ADMET properties
pkpd-predict -i compounds.csv -o predictions.csv

# Fit PK model
pkpd-fit -d pk_data.csv -s human -m 2compartment

# Simulate NOAEL
pkpd-simulate -m fitted_model.pkl -d dosing.yaml

# Calculate therapeutic index
pkpd-ti -p predictions.csv

# Score compounds
pkpd-score -c compounds.csv -t targets.yaml

🏗️ Architecture

src/pkpd/
├── admet/           # ADMET prediction (CL, F, t½, DILI, hERG)
├── models/          # PK/PD model library (PySB)
├── fit/             # Fitting pipeline (SciPy alternatives)
├── simulate/        # Virtual population simulation
├── ti/              # Therapeutic index calculators
├── design/          # Compound scoring and design
├── io/             # Data I/O and validation
├── reporting/       # Plots, tables, reports
└── cli/             # Command-line interface

configs/
├── species/         # Human, rodent, dog, NHP configs
├── dosing/          # 14d, 1m, 9m regimens
├── tox_thresholds/  # QTc, marrow, DILI thresholds
└── fitting/         # Fitting options and priors

🔧 Libraries Used

Core Scientific Stack

• SciPy - ODE solving, optimization, scientific computing
• NumPy - Numerical computing
• Pandas - Data manipulation
• Matplotlib/Seaborn - Visualization

Specialized Libraries

• RDKit - Cheminformatics and molecular descriptors
• PySB - Systems biology modeling
• ADMET-AI - ADMET property prediction

No Complex Dependencies

• ❌ No AMICI (replaced with SciPy ODE solvers)
• ❌ No pyPESTO (replaced with SciPy optimization)
• ❌ No PyTDC (replaced with direct data loading)

📊 Usage Examples

1. ADMET Prediction

from pkpd.admet import ADMETPredictor
predictor = ADMETPredictor()
predictions = predictor.predict_batch(smiles_list, compound_ids)

2. PK/PD Modeling

from pkpd.models import TwoCompartmentPK
pk_model = TwoCompartmentPK(species='human')
pk_model.scale_to_species(70.0)

3. Cross-species Simulation

from pkpd.simulate import SimulationEngine
engine = SimulationEngine()
noael = engine.simulate_noael(model, dosing_regimen)

4. Compound Design

from pkpd.design import CompoundDesigner
designer = CompoundDesigner()
scores = designer.score_compounds(compounds, target_profiles)

🎯 Goals Achieved

✅ Goal 1: Human PK/TI Prediction

• Predicted clearance, volume, half-life, bioavailability
• Assessed QTc, bone marrow, and DILI risks
• Generated risk-based compound ranking

✅ Goal 2: Cross-species PK/PD Models & NOAEL

• Built 2-compartment PK models for 4 species
• Applied allometric scaling for cross-species translation
• Estimated NOAEL for 14d/1m/9m regimens
• Generated safety margin analysis

✅ Goal 3: Compound Design & Optimization

• Scored compounds against target profiles
• Performed multi-objective optimization
• Applied global optimization (differential evolution)
• Generated design recommendations

🚀 Production Deployment

Docker

FROM python:3.11-slim
# Simple deployment - no complex compilation needed

GKE

• Kubernetes manifests in k8s/
• Resource management and health checks
• Persistent storage for data

Benefits

• ✅ Simpler deployment (no complex compilation)
• ✅ Faster builds (no AMICI/pyPESTO compilation)
• ✅ Easier maintenance (standard Python libraries)
• ✅ Production ready (works in any environment)

📁 Repository Structure

PKPD/
├── src/pkpd/           # Core platform code
├── configs/            # Configuration files
├── scripts/            # CLI scripts
├── notebooks/          # Jupyter notebooks
├── tests/              # Unit tests
├── k8s/                # Kubernetes manifests
├── data/               # Data directory (git-ignored)
├── models_artifacts/   # Model artifacts (git-ignored)
├── Dockerfile          # Production Docker image
├── environment.yml     # Conda environment
└── pyproject.toml      # Python dependencies

🤝 Contributing

1. Fork the repository
2. Create a feature branch
3. Make your changes
4. Add tests
5. Submit a pull request

📄 License

MIT License - see LICENSE file for details.

🎉 Status

✅ PRODUCTION READY

All three goals achieved using working libraries and SciPy alternatives. No complex dependencies needed - the platform is fully functional and ready for production deployment!