mindscience/MindSPONGE

• MindSpore SPONGE
• • Introduction
  • Latest News 📰
  • Coming Soon 🚀
  • Quick Start
  • • Protein Multimer Structure Prediction
    • A simple example for molecular dynamics
  • Installation
  • • Version dependency
    • Hardware
    • pip install
    • source code install
    • API
  • Community
  • • CO-CHAIR
    • Special Interesting Group 🏠
    • Core Contributor 🧑‍🤝‍🧑
    • Cooperative Partner
  • Contribution Guide
  • License

ENGLISH | 简体中文

MindSpore SPONGE

Introduction

MindSpore SPONGE(Simulation Package tOwards Next GEneration molecular modelling) is a toolkit for Computational Biology based on AI framework MindSpore，which supports MD, folding and so on. It aims to provide efficient AI computational biology software for a wide range of scientific researchers, staff, teachers and students.

Latest News 📰

• 🔥Top open source internship task has been released! Everyone is welcome to claim it~
• 🔥2023.1.31 MindSPONGE version 1.0.0-alpha is released. The documents are available on Scientific Computing MindSPONGE module on MindSpore website
• 2022.8.23 Paper "Few-Shot Learning of Accurate Folding Landscape for Protein Structure Prediction" is preprinted in arxiv, Please refer to Paper
• 2022.8.11—2022.8.15 MindSpore SPONGE SIG Summer School, replay
• 2022.07.18 Paper "SPONGE: A GPU-Accelerated Molecular Dynamics Package with Enhanced Sampling and AI-Driven Algorithms"is published in Chinese Journal of Chemistry. Please refer to paper and codes
• 2022.07.09 MEGA-Assessment wins CAMEO-QE monthly 1st
• 2022.06.27 Paper "PSP: Million-level Protein Sequence Dataset for Protein Structure Prediction" is preprinted in arxiv. Please refer to Paper and codes.
• 2022.04.21 MEGA-Fold wins CAMEO-3D monthly 1st. Related News

Coming Soon 🚀

• Everything is coming soon, don't worry~

Quick Start

Protein Multimer Structure Prediction

import os
import stat
import pickle
from mindsponge import PipeLine
from mindsponge.common.protein import to_pdb_v2, from_prediction_v2

cmd = "wget https://download.mindspore.cn/mindscience/mindsponge/Multimer/examples/6T36.pkl"
os.system(cmd)

pipe = PipeLine(name="Multimer")
pipe.set_device_id(0)
pipe.initialize("predict_256")
pipe.model.from_pretrained()
f = open("./6T36.pkl", "rb")
raw_feature = pickle.load(f)
f.close()
final_atom_positions, final_atom_mask, confidence, b_factors = pipe.predict(raw_feature)
unrelaxed_protein = from_prediction_v2(final_atom_positions,
                                       final_atom_mask,
                                       raw_feature["aatype"],
                                       raw_feature["residue_index"],
                                       b_factors,
                                       raw_feature["asym_id"],
                                       False)
pdb_file = to_pdb_v2(unrelaxed_protein)
os.makedirs('./result/', exist_ok=True)
os_flags = os.O_RDWR | os.O_CREAT
os_modes = stat.S_IRWXU
pdb_path = './result/unrelaxed_6T36.pdb'
with os.fdopen(os.open(pdb_path, os_flags, os_modes), 'w') as fout:
    fout.write(pdb_file)
print("confidence:", confidence)

A simple example for molecular dynamics

import numpy as np
from mindspore import context
from mindsponge import Sponge
from mindsponge import Molecule
from mindsponge import ForceFieldBase
from mindsponge import DynamicUpdater
from mindsponge.potential import BondEnergy, AngleEnergy
from mindsponge.callback import WriteH5MD, RunInfo
from mindsponge.function import VelocityGenerator
from mindsponge.control import LeapFrog

context.set_context(mode=context.GRAPH_MODE, device_target="GPU")

system = Molecule(
    atoms=['O', 'H', 'H'],
    coordinate=[[0, 0, 0], [0.1, 0, 0], [-0.0333, 0.0943, 0]],
    bond=[[[0, 1], [0, 2]]],
)

bond_energy = BondEnergy(
    index=system.bond,
    force_constant=[[345000, 345000]],
    bond_length=[[0.1, 0.1]],
)

angle_energy = AngleEnergy(
    index=[[1, 0, 2]],
    force_constant=[[383]],
    bond_angle=[[109.47 / 180 * np.pi]],
)

energy = ForceFieldBase(energy=[bond_energy, angle_energy])

velocity_generator = VelocityGenerator(300)
velocity = velocity_generator(system.coordinate.shape, system.atom_mass)

opt = DynamicUpdater(
    system,
    integrator=LeapFrog(system),
    time_step=1e-3,
    velocity=velocity,
)

md = Sponge(system, energy, opt)

run_info = RunInfo(10)
cb_h5md = WriteH5MD(system, 'test.h5md', save_freq=10, write_velocity=True, write_force=True)

md.run(1000, callbacks=[run_info, cb_h5md])

More Cases：👀

• Protein Relaxation
• Protein Structure Prediction MEGA-Fold
• Protein Structure Assessment MEGA-Assessment
• Evolution Engine MEGA-EvoGen
• Function based protein design (TO BE DONE)
• Structure based protein design (TO BE DONE)
• Protein function prediction (TO BE DONE)
• Molecular representation model (TO BE DONE)

Installation

Version dependency

Due to the dependency between MindSPONGE and MindSpore, please follow the table below and install the corresponding MindSpore version from MindSpore download page.

MindSPONGE Version	Branch	MindSpore Version	Python Version
1.0.0	master	>=2.0.0-alpha	>=3.7

pip install -r requirements.txt

Hardware

Hardware	OS	Status
Ascend 910	Ubuntu-x86	✔️
	Ubuntu-aarch64	✔️
	EulerOS-aarch64	✔️
	CentOS-x86	✔️
	CentOS-aarch64	✔️
GPU CUDA 10.1	Ubuntu-x86	✔️

• CUDA>=10.1
• Ubuntu>=16.04

pip install

pip install mindsponge-[gpu|ascend]

The version of mindsponge installed by pip corresponds to the r0.2.0-alpha branch code. The code can be downloaded using the following instruct.

git clone -b r0.2.0-alpha https://gitee.com/mindspore/mindscience.git

source code install

git clone https://gitee.com/mindspore/mindscience.git
cd {PATH}/mindscience/MindSPONGE

• Ascend backend

bash build.sh -e ascend

• GPU backend

Enable c if you want to use Cybertron.

Enable t if you want to use traditional MD.

export CUDA_PATH={your_cuda_path}
bash build.sh -e gpu -j32 -t on -c on

• Install whl package

cd {PATH}/mindscience/MindSPONGE/output
pip install mindsponge_ascend*.whl # Ascend
pip install mindsponge-gpu*.whl # GPU
pip install cybertron*.whl # if "-c on" is used

API

For details about MindSPONGE APIs, please refer to API pages.

Community

CO-CHAIR

                    
Shenzhen Bay Laboratory Yi Isaac Yang     Chang Ping Laboratory Jun Zhang     Chang Ping Laboratory Sirui Liu

Special Interesting Group 🏠

MindSpore SPONGE SIG (Special Interesting Group) is a team composed of a group of people who are interested and have a mission to make achievements in the field of AI × biological computing.

MindSpore SPONGE SIG group provides efficient and easy-to-use AI computational biology software for researchers, teachers and students, and provides a platform for people with strong abilities or strong interests in this field to communicate and cooperate together.

At present, the SIG group has six core teachers. After members joining the SIG group, our teachers will lead the team to carry out scientific research and develop the software function development. Of course, members are also welcome to do research on their own topics using MindSPONGE.

In the SIG group, we will hold various activities, including summer school, public lecture, technology communication meeting and other large-scale activities. Small-scale activities like weekly meetings, blogs writing will also be held in the group. By joining the activities, there will be lots of chances to communicate with our experts. During the summer school program ended on August 15th, we invited 13 teachers to have a five-day lecture mainly including three themes of MindSpore basics, molecular dynamics and advanced AI × Science courses. You can get the replay here.

In the SIG group, we will also release the public intelligence task and open source internship task, welcome everyone to claim it.

If you want to join us and become a member of our group, please send your resume to dingyahao@huawei.com, we are always looking forward to your arrival.

Core Contributor 🧑‍🤝‍🧑

• Yi Qin Gao Research Group: Yi Isaac Yang，Jun Zhang，Sirui Liu，Yijie Xia，Diqing Chen，Yu-Peng Huang

Cooperative Partner

   

Contribution Guide

• Please click here to see how to contribute your code:Contribution Guide

License

Apache License 2.0