github.com/PKU-YuanGroup/TaxDiff @main

TaxDiff: Taxonomic-Guided Diffusion Model for Protein Sequence Generation

[](https://arxiv.org/abs/2402.17156) [](https://github.com/HowardLi1984/ECDFormer/blob/main/LICENSE) [](https://github.com/HowardLi1984/ECDFormer/blob/main/LICENSE) [](https://github.com/HowardLi1984/ECDFormer/blob/main/DATASET_LICENSE)

If you like our project, please give us a star ⭐ on GitHub for latest update.

The official code for "TaxDiff: Taxonomic-Guided Diffusion Model for Protein Sequence Generation". Here we publish the inference code of TaxDiff. The training code & Protein sequence with Taxonomic lables dataset will be released after our paper is accepted.

💡 I also have other AI for Science projects that may interest you ✨. > [**ProLLaMA: A Protein Large Language Model for Multi-Task Protein Language Processing**](https://github.com/Lyu6PosHao/ProLLaMA) >Liuzhenghao Lv, Zongying Lin, Li Hao, Yuyang Liu, Jiaxi Cui, Calvin Yu-Chian Chen, Li Yuan, Yonghong Tian [](https://github.com/Lyu6PosHao/ProLLaMA) [](https://arxiv.org/abs/2402.16445) ## 😮 Highlights ### 💡 Protein sequences Generation Model - To the best of our knowledge, our TaxDiff is **the first controllable protein generation model** utilizing guidance from taxonomies. ### 🔥 Diffusion-based Framework - TaxDiff proposes a **taxonomic-guided framework** that fits all diffusion-based protein design models. We also propose the patchify attention mechanism for better protein design. ### ⭐ Excellent performance - Experiments demonstrate that our TaxDiff achieves **state-of-the-art results** in both taxonomic-guided controllable and unconditional protein sequence generation, excelling in structural modeling scores and sequence consistency. ## 🚀 Main Results More detailed results can be found in our paper. ### Unconditional Generation ### Controllable Generation ## 📖 Data Preparation For inference, please download from [HuggingFace](https://huggingface.co/linzy19/TaxDiff/tree/main). Unzip it and put the [ckpt](https://huggingface.co/linzy19/TaxDiff/tree/main) into the folder ckpt/

ckpt/0012802_eval.ckpt

Our dataset can download from [HuggingFace](https://huggingface.co/linzy19/TaxDiff/tree/main).

uniref50_200_256_clean_taxnomic_family_tid__filter_layer6.fasta

We will release protein sequences with taxonmic labels for training procedure once our paper is accepted. If you want to select a specific protein taxonomic for your research, you need to first find his corresponding tax-id in the [data_reader/Taxonnmic_classfication.xlsx](https://github.com/Linzy19/TaxDiff/blob/main/data_reader/Taxonnmic_classfication.xlsx), and then modify protein class lables in the [sample_protein.py](https://github.com/Linzy19/TaxDiff/blob/main/sample_protein.py).

class_lables = torch.randint(low=1, high=int(23427), size=(1,num))

## 🛠️ Requirements and Installation * Python == 3.10 * Pytorch == 2.2.0 * Torchvision == 0.17.0 * CUDA Version == 12.0 * Install required packages:

git clone git@[github.com/Linzy19/TaxDiff.git]
cd TaxDiff
pip install -r requirements.txt

## 🗝️ Inferencing The inferencing instruction is in [sample_protein.py](sample_protein.py).

python sample_protein.py --model DiT-pro-12-h6-L16 --cuda-num cuda:0 --num 500

## ✏️ Citation If you find our paper and code useful in your research, please consider giving a star :star: and citation :pencil:.

@article{zongying2024taxdiff,
  title={TaxDiff: Taxonomic-Guided Diffusion Model for Protein Sequence Generation},
  author={Zongying, Lin and Hao, Li and Liuzhenghao, Lv and Bin, Lin and Junwu, Zhang and Yu-Chian, Chen Calvin and Li, Yuan and Yonghong, Tian},
  journal={arXiv preprint arXiv:2402.17156},
  year={2024}
}