mindspore/model_zoo/research/hpc/sponge

• SPONGE Example
• • Description
  • Dataset
  • Environment Requirements
  • Quick Start
  • Script Description
  • • Script and Sample Code
    • Training Process
  • Result
  • Model Description
  • • Evaluation Performance
  • ModelZoo HomePage

SPONGE Example

• Description
• Dataset
• Environment Requirements
• Quick Start
• Script Description
  • Script and Sample Code
  • Training Process
• Model Description
  • Evaluation Performance
• Result
• ModelZoo Homepage

Description

SPONGE in MindSpore is a high-performance and modularized molecular dynamics simulation library developed by Yiqin Gao group (Peking University and Shenzhen Bay Laboratory) and MindSpore team at Huawei Company. It can efficiently simulate traditional molecular dynamics tasks based on the “graph-kernel-fusion” and “automatic parallelization” features of MindSpore. In the meanwhile, it utilizes the automatic differentiation feature of MindSpore, and introduces machine learning methods, such as neural network, into traditional molecular simulation, achieving methodological inventions.

This example demonstrates how to perform high-performance molecular dynamics simulations with the built-in SPONGE module of MindSpore on GPU.

Dataset

There are three inputs for the example, property file NVT_290_10ns.in, topology file WATER_ALA.parm7 and coordinates file WATER_ALA_350_cool_290.rst7, respectivelly.

Topology file and coordinates file can be generated by tleap in AmberTools (link). For more details, please refer to:

• SPONGE Tutorial

Environment Requirements

• Hardware: GPU
  • Prepare hardware environment with GPU processor.
• Framework
  • MindSpore
• For more information, please check the resources below:
  • MindSpore Tutorials
  • MindSpore Python API

Quick Start

After installing MindSpore via the official website, you can start running as follows:

python main.py --i /path/NVT_290_10ns.in \
               --amber_parm /path/WATER_ALA.parm7 \
               --c /path/WATER_ALA_350_cool_290.rst7 \
               --o /path/ala_NVT_290_10ns.out

path is the path which stores input files.

Script Description

Script and Sample Code

├── sponge
    ├── README.md
    ├── main.py                                  # launch Simulation for SPONGE
    ├── src
        ├── bond.py                              # bond module in SPONGE
        ├── angle.py                             # angle module in SPONGE
        ├── dihedral.py                          # dihedral module in SPONGE
        ├── nb14.py                              # nb14 module in SPONGE
        ├── Langevin_Liujian_md.py               # Langevin_Liujian_md module in SPONGE
        ├── lennard_jones.py                     # lennard_jones module in SPONGE
        ├── md_information.py                    # save md information module in SPONGE
        ├── neighbor_list.py                     # neighbor_list module in SPONGE
        ├── particle_mesh_ewald.py               # particle_mesh_ewald module in SPONGE
        ├── simulation.py                # SPONGE simulation

Training Process

python main.py --i /path/NVT_290_10ns.in \
               --amber_parm /path/WATER_ALA.parm7 \
               --c /path/WATER_ALA_350_cool_290.rst7 \
               --o /path/ala_NVT_290_10ns.out

Training result will be stored in the specified file, which ends with ".out".

Result

After training，the results in ala_NVT_290_10ns.out are：

_steps_ _TEMP_ _TOT_POT_ENE_ _BOND_ENE_ _ANGLE_ENE_ _DIHEDRAL_ENE_ _14LJ_ENE_ _14CF_ENE_ _LJ_ENE_ _CF_PME_ENE_
      1 293.105   -6117.709   1204.406       7.096          4.491      3.456     44.018 1372.488    -8753.664
   ...

There are sorts of energy in the output, steps (steps), temperature (TEMP), total energy (TOT_POT_E), bond energy (BOND_ENE), angle energy (ANGLE_ENE), dihedral energy (DIHEDRAL_ENE), non bond enrgy, includes Coulomb force (14CF_ENE) and Leonard-Jones energy (14LJ_ENE), Van der Waals energy (LJ_ENE) and Coulomb force in PME (CF_PME_ENE).

Model Description

Evaluation Performance

Parameters	GPU
Resource	GPU(Tesla V100 SXM2), Memory 16G
uploaded Date
MindSpore Version	1.2
Training Parameters	step=1
Outputs	numpy file
Speed	16.7 ms/step
Total time	10.7 s
Scripts	Link

ModelZoo HomePage

Please check the official homepage.