PandaKinetics

Multi-Scale Structure-Kinetics Simulator for Drug Design

Overview

PandaKinetics is a GPU-accelerated toolkit for predicting drug binding kinetics, residence times, and kinetic selectivity using AI-enhanced molecular dynamics simulations.

Features

• Enhanced Docking: Multi-site ensemble docking with conformational diversity
• AI-Powered Barriers: Machine learning prediction of transition state energies
• Kinetic Monte Carlo: GPU-accelerated simulation of binding/unbinding kinetics
• Kinetic Selectivity: Prediction of selectivity based on residence times
• Comprehensive Analysis: Statistical analysis and visualization tools

Installation

Requirements

• Python 3.8+
• CUDA-capable GPU
• CUDA Toolkit 12.0+

Install from PyPI

bash pip install pandakinetics

Install from source

bash git clone https://github.com/pandakinetics/pandakinetics.git cd pandakinetics pip install -e .

Quick Start

Basic Usage

```python from pandakinetics import KineticSimulator

Initialize simulator

simulator = KineticSimulator( temperature=310.0, # Physiological temperature n_replicas=16, # Parallel simulations )

Predict kinetics

results = simulator.predictkinetics( proteinpdb="1ABC", # PDB ID or file path ligand_smiles="CCO" # Ethanol as example )

Access results

print(f"Association rate: {results.kon:.2e} M⁻¹s⁻¹") print(f"Dissociation rate: {results.koff:.2e} s⁻¹") print(f"Residence time: {results.residence_time:.2e} s") ```

Command Line Interface

```bash

Predict kinetics

pandakinetics predict --protein 1ABC --ligand "CCO" --output results/

Visualize results

pandakinetics visualize --results-file results/kinetic_results.json

Benchmark GPU

pandakinetics benchmark ```

Advanced Usage

```python from pandakinetics import KineticSimulator, BarrierPredictor

Custom barrier predictor

barrierpredictor = BarrierPredictor( modelpath="custommodel.pt", hiddendim=256 )

Advanced simulator configuration

simulator = KineticSimulator( temperature=310.0, nreplicas=32, maxsimulationtime=1e-2, # 10 ms barrierpredictor=barrier_predictor )

Predict with custom binding sites

binding_sites = [ {'center': [10.0, 15.0, 20.0], 'radius': 10.0}, {'center': [30.0, 25.0, 10.0], 'radius': 8.0} ]

results = simulator.predictkinetics( proteinpdb="protein.pdb", ligandsmiles="complexmoleculesmiles", bindingsites=bindingsites, referenceligands=["referencesmiles1", "referencesmiles2"] )

Analyze selectivity

for refligand, selectivity in results.kineticselectivity.items(): print(f"Selectivity vs {ref_ligand}: {selectivity:.2f}") ```

Key Components

1. Enhanced Docking Engine

• Multi-conformer generation
• Ensemble docking across multiple binding sites
• Pose clustering and filtering
• GPU-accelerated scoring

2. AI Barrier Predictor

• E(3)-equivariant neural networks
• Transition state energy prediction
• Physics-informed constraints
• Pre-trained on MD simulation data

3. Kinetic Monte Carlo Simulator

• Parallel replica simulations
• Transition rate calculations
• Binding/unbinding event detection
• Statistical analysis

4. Analysis and Visualization

• Kinetic parameter estimation
• Confidence interval calculation
• Pathway analysis
• Interactive plotting

Performance

PandaKinetics is optimized for GPU acceleration:

• Docking: 100+ poses in seconds
• Barrier Prediction: 1000+ transitions per second
• Monte Carlo: Million+ steps per second (parallel)
• Memory: Efficient GPU memory management

Applications

Drug Discovery

• Lead optimization for residence time
• Kinetic selectivity design
• ADMET property prediction

Research

• Binding mechanism elucidation
• Allosteric pathway analysis
• Structure-kinetics relationships

Validation

PandaKinetics has been validated against: - Experimental SPR data - Literature kinetic parameters - Benchmark datasets

Citation

If you use PandaKinetics in your research, please cite:

@article{pandakinetics2025, title={PandaKinetics: Multi-Scale Structure-Kinetics Simulator for Drug Design}, author={Pritam Kumar Panda}, year={2025} }

License

MIT License - see LICENSE file for details.

Support

• Documentation: https://pritampanda15.readthedocs.io
• Issues: https://github.com/pritampanda15/pandakinetics/issues
• Discussions: https://github.com/pritampanda15/pandakinetics/discussions

Contributing

We welcome contributions! Please see CONTRIBUTING.md for guidelines.

Acknowledgments

• Built on PyTorch, RDKit, OpenMM, and Biotite
• Inspired by advances in AI-accelerated MD