Skip to content

Open-Protein is an open source pre-training platform that supports multiple protein pre-training models and downstream tasks.

License

Notifications You must be signed in to change notification settings

HICAI-ZJU/OpenProtein

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

5 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Open-Protein

Open-Protein is an open source pre-training platform that supports multiple protein pre-training models and downstream tasks.

Qucik Start

This repo is tested on Python 3.6 - 3.9 and PyTorch 1.5.0+ (PyTorch 1.5.0+ for examples).

When you install Open-Protein, we do not force the installation of pytorch dependencies, you are free to choose a version of Pytorch 1.5 or higher.

With Pip

First you need to install PyTorch and torch-scatter, where the version number of torch should be greater than 1.5.0, and the version number of torch-scatter should be greater than 2.0.8. Please refer to PyTorch installation page regarding the specific install command for your platform.

When PyTorch has been installed, Open-Protein can be installed using pip as follows:

pip install openprotein

From source

First you need to install PyTorch and torch-scatter, where the version number of torch should be greater than 1.5.0, and the version number of torch-scatter should be greater than 2.0.8. Please refer to PyTorch installation page regarding the specific install command for your platform.

When PyTorch has been installed, you can install from source by cloning the repository and running:

git clone https://github.com/HIC-AIX/OpenProtein.git
cd open-protein
pip install .

When you update the repository, you should upgrade the transformers installation and its dependencies as follows:

git pull
pip install --upgrade .

Content

Dataset

You can download the corresponding datasets according to the downstream tasks. The links for each dataset are listed below.

Name url
Uniref Download
EC Download
Flip Download
Tape Download
Go Download

Model

model introduction paper github weight
Esm1b A 650M-parameter protein language model. Genome-wide prediction of disease variants with a deep protein language rmrao/esm-1: Evolutionary Scale Modeling (esm): Pretrained language models for proteins https://dl.fbaipublicfiles.com/fair-esm/models/esm_msa1b_t12_100M_UR50S.pt
GearNet Geometry-Aware Relational Graph Neural Network. GearNet: Stepwise Dual Learning for Weakly Supervised Domain Adaptation GitHub - DeepGraphLearning/GearNet: GearNet and Geometric Pretraining Methods for Protein Structure Representation Learning, ICLR Pre-Trained Models for GearNet
ProteinBert ProtBert is based on Bert model which pretrained on a large corpus of protein sequences in a self-supervised fashion. ProtTrans: Towards Cracking the Language of Life’s Code Through Self-Supervised Learning GitHub - agemagician/ProtTrans: ProtTrans is providing state of the art pretrained language models for proteins. ProtTrans was trained on thousands of GPUs from Summit and hundreds of Google TPUs using Transformers Models. Rostlab/prot_bert · Hugging Face

Downstream Task

Task introduction paper dataset
EC a numerical classification scheme for enzymes, based on the chemical reactions they catalyze. As a system of enzyme nomenclature, every EC number is associated with a recommended name for the corresponding enzyme-catalyzed reaction. Structure-based protein function prediction using graph convolutional networks https://users.flatironinstitute.org/~renfrew/DeepFRI_data/PDB-EC.tar.gz
FLIP a benchmark for function prediction to encourage rapid scoring of representation learning for protein engineering. FLIP: Benchmark tasks in fitness landscape inference for proteins https://github.com/J-SNACKKB/FLIP/tree/main/splits
TAPE a set of five biologically relevant semi-supervised learning tasks spread across different domains of protein biology. Evaluating Protein Transfer Learning with TAPE https://github.com/songlab-cal/tape#lmdb-data
GO GO (gene ontology) is a database established by the Gene Ontology Consortium. It aims to establish a language vocabulary standard that is applicable to various species, defines and describes the functions of genes and proteins, and can be updated with the deepening of research. GO is one of a variety of biological ontology languages, which provides a three-layer system definition method to describe the function of gene products. Gene Ontology: tool for the unification of biology https://users.flatironinstitute.org/~renfrew/DeepFRI_data/PDB-GO.tar.gz

Usage

Data Process

  • Take the EC dataset as an example.
 # 1. Download ec dataset.
wget -O ./data/PDB-EC.tar.gz -c https://users.flatironinstitute.org/~renfrew/DeepFRI_data/PDB-EC.tar.gz

 # 2. Create a new folder for the ec dataset.
mkdir ./data/ori_ec

 # 3. Unzip the ec dataset to the new folder.
tar -zxvf PDB-EC.tar.gz -C ./data/ori_ec

 # 4. Executing dataset processing files.
python conver_to_lmdb.py ec -p ./data/ori_ec -o ./data/ec

Feature Extraction

from openprotein import Esm1b, Esm1bConfig
from openprotein.data import MaskedConverter, Alphabet

seq = 'RLQIEAIVEGFTQMKTDLEKEQRSMASMWKKREKQIDKVLLNTTYMYGSIKGIAGNAVQTVSLLELPVDENGEDE'

converter = MaskedConverter.build_convert()
alphabet = Alphabet.build_alphabet()

args = Esm1bConfig()

origin_tokens, masked_tokens, target_tokens = converter(seq)
model = Esm1b(args, alphabet)

feature = model(masked_tokens)

Downstream Task

from openprotein import Esm1b, Esm1bConfig
from openprotein.data import MaskedConverter, Alphabet, TaskConvert
from openprotein.task import ProteinFunctionDecoder

seq = 'RLQIEAIVEGFTQMKTDLEKEQRSMASMWKKREKQIDKVLLNTTYMYGSIKGIAGNAVQTVSLLELPVDENGEDE'

converter = MaskedConverter.build_convert()
alphabet = Alphabet.build_alphabet()
args = Esm1bConfig(checkpoint_path="./resources/esm1b/esm1b_t33_650M_UR50S.pt")

origin_tokens, masked_tokens, target_tokens = converter(seq)
model = Esm1b.load(args, alphabet).eval()

feature = model(masked_tokens)

converter = TaskConvert(alphabet)
protein_function_decoder = ProteinFunctionDecoder(args.embed_dim, args.class_num)
outputs = protein_function_decoder(feature, seq)
print(outputs)

License

This source code is licensed under the OSL 3.0 license found in the LICENSE file in the root directory of this source tree

About

Open-Protein is an open source pre-training platform that supports multiple protein pre-training models and downstream tasks.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages