README.md

ORI

---

English | 简体中文

---

About

This repository is the ORI protein design warehouse of Tencent AI For Life Sciences Lab, including projects such as protein generation, protein attribute prediction and protein basic model reinforcement learning.

De Novo Design of Functional Proteins with ORI
Bin He, Chenchen Qin...Jianhuayao
Paper: https://arxiv.org/abs/xxxxx

TX-L2

TX-L6

Hen Lysozyme

Main Models

Project	Model	Dataset	Description
Protein Generation	ORI-PGM-1B	Uniref50,PDB	ORI protein generation 1B model
	ORI-PGM-3B	Uniref50, PDB	ORI protein generation 3B model
Protein Discriminator	USM-100M	Uniclust30, Uniref50	USM 100M mono sequence and msa foundation model
	USM-100M-Solubility	Solubility dataset	USM 100M solubility prediction model
	USM-100M-Thermostablility	Thermostablility dataset	USM 100M thermostablility prediction model
	USM-100M-SignalP	Signal peptide dataset	USM 100M signal pepetide prediction model
xfold	USMFold-100M(will be released soon)	Uniref50, PDB	Superfast protein folding prediction model based on USM
	ESMFold	Uniref50, PDB, AFDB	Optimized ESMFold in ori program
RLWF			Reinforcement Learning from Wet-lab Feedback

you can download pre-trained weights with following link:

• Tencent Weiyun Drive
• Google Drive
• Zenodo

Getting Started

Install

Requirements

• System: Linux and MacOS
• Python 3.8 and above
• Pytorch 2.0.0 and above, no more than 2.4.0
• If you use Nvidia GPU, make sure the memory is greater than 8G

To install

You can install the package with the following command line. For other installation methods and options, please refer to INSTALL.md.

# install miniconda
wget -O minicnda3.sh https://repo.anaconda.com/miniconda/Miniconda3-py39_24.5.0-0-Linux-x86_64.sh
# specific miniconda install path
CONDA_PATH=/miniconda
bash minicnda3.sh -b -p ${CONDA_PATH}
rm minicnda3.sh
# init environment
conda init
# download code
git clone https://github.com/TencentAI4S/ori.git
cd ori
conda env create -n ori -f environment.yml
conda activate ori

Download Model Weights(Optional)

if you want to test model offline, please download model weights to "~/.cache/torch/hub/checkpoints" first.

• Tencent Weiyun Drive
• Google Drive
• Zenodo

1. Protein Generation

Lysozyme Generation

prompt='<Glucosaminidase><temperature90><:>'
python projects/progen/generate_protein.py -p ${prompt} -n 5

Enzyme Generation

prompt='<EC:3.1.1.101><temperature90><:>'
python projects/progen/generate_protein.py -p ${prompt} -n 5

Multifunctional Enzyme Generation

prompt='<EC:3.2.1.14><EC:3.2.1.17><temperature60><:>'
python projects/progen/generate_protein.py -p ${prompt} -n 5

2. Protein Solubility Prediction

python projects/prodiscriminator/predict_solubility.py -i projects/prodiscriminator/data/solubility_demo.fasta

3. Protein Thermostablility Prediction

python projects/prodiscriminator/predict_thermostability.py -i projects/prodiscriminator/data/thermostability_demo.fasta

4. Signal Peptide Prediction

python projects/prodiscriminator/predict_signal_peptide.py -i projects/prodiscriminator/data/signalp_demo.fasta

5. Protein Fold Prediction with USMFold

python projects/xfold/usmfold_predict.py -i projects/xfold/data/test.fasta

Citation

If you use this codebase, or otherwise find our work valuable, please cite ori:

@article{ori,
  title={De Novo Design of Functional Proteins with ORI},
  author={Bin He,Chenchen Qin...Jianhuayao},
  journal={arXiv preprint arXiv:xxx},
  year={2025}
}