2025Molecular Dynamics Simulation

3D Protein

View on GitHub↗
Built with
01 / Overview

A GPU-accelerated coarse-grained molecular dynamics framework for protein structure simulation using NVIDIA Warp and interactive 3D visualization with Polyscope.

Understanding protein dynamics is fundamental to drug discovery and structural biology. This project implements a complete coarse-grained molecular dynamics pipeline using NVIDIA Warp for differentiable, GPU-accelerated physics simulation and Polyscope for real-time 3D visualization. The framework employs a one-bead-per-residue coarse-grained representation to simulate protein motion via overdamped Langevin dynamics, incorporating harmonic bond forces, Lennard-Jones van der Waals interactions, and Coulomb electrostatics. Using the tumor suppressor protein p53 (PDB: 1TUP) as a demonstration case, the system achieves real-time simulation performance while maintaining modular architecture separating physics engine from visualization components.
02 / Process
01

Structure Parsing & System Initialization

Parsed protein structure files in CIF format and constructed coarse-grained representation with one bead per amino acid. Analyzed chain topology, bond connectivity, and residue properties (hydrophobic, polar, charged) to initialize the simulation system with appropriate force field parameters.

02

GPU-Accelerated Simulation

Implemented overdamped Langevin dynamics using NVIDIA Warp with custom GPU kernels for force calculations and time integration. Applied harmonic bonds for chain connectivity, Lennard-Jones 12-6 potential for van der Waals interactions, and Coulomb potential for electrostatics between charged residues. Achieved real-time performance through CUDA acceleration with automatic CPU fallback.

03

Interactive Visualization & Export

Built Polyscope-based 3D visualization supporting interactive viewing, before/after structure comparison, and animated trajectory playback. Generated USD format output for integration with modern 3D graphics pipelines. Modular design separates simulation engine from visualization for independent research use.

03 / Impact
  • ◆

    Implemented GPU-accelerated coarse-grained molecular dynamics using NVIDIA Warp with custom CUDA kernels for harmonic bond forces, Lennard-Jones potentials, and Langevin integration, achieving real-time simulation performance on both CUDA and CPU backends.

  • ◆

    Developed modular architecture separating physics simulation engine from Polyscope visualization layer, enabling independent component usage, trajectory export to USD format, and research reproducibility.

  • ◆

    Created interactive visualization system with before/after conformational comparison, animated trajectory playback, and chain topology analysis using tumor suppressor protein p53 (PDB: 1TUP) as demonstration case.

"

"This project bridges the gap between computational physics and molecular biology, demonstrating how modern GPU programming can accelerate our understanding of protein dynamics and contribute to advances in structural biology and drug design."