Welcome to the official repository for the ESM-AA (ESM-All Atom) model (paper link). This repository contains the source code and usage instructions for the ESM-AA model, developed based on the official ESM codes by FAIR. You can find the original ESM repository here.

To set up the necessary environment for using the ESM-AA model, follow these steps:

1. Clone the repository:

   git clone https://github.com/zhengkangjie/ESM-AA.git
   cd ESM-AA

2. Create and activate the conda environment:

   conda env create -f environment.yml
   conda activate esmfold

3. Install the package:

   python setup.py install

We have made available checkpoints for the ESM-AA model. There is one version available now and we will release more checkpoints of ESM-AA with larger scale soon:

• ESM-AA 35M with Regression Weight: Download from this link.

Currently, we provide an example for the unsupervised contact prediction task using the same settings as the original ESM model. You can find the example code in the examples directory.

Here is an example to load ESM-AA:

import esm

# Load model and alphabet
esm_aa, esm_aa_alphabet = esm.pretrained.esm_aa_t12_35M('path/to/checkpoint')

We plan to release code for protein-molecular tasks soon. Stay tuned for updates and more functionalities!

If you find the models useful in your research, please cite our paper:

@article{zheng2024multi,
  title={Multi-Scale Protein Language Model for Unified Molecular Modeling},
  author={Zheng, Kangjie and Long, Siyu and Lu, Tianyu and Yang, Junwei and Dai, Xinyu and Zhang, Ming and Nie, Zaiqing and Ma, Wei-Ying and Zhou, Hao},
  journal={bioRxiv},
  pages={2024--03},
  year={2024},
  publisher={Cold Spring Harbor Laboratory}
}